LLeMbas Desktop 1.0.0-beta.6: CLI 2.0.0-beta.6, the look and words of LLeMbas 2.0.0-beta.8

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

.github/workflows/ci.yml

@@ -1,7 +1,8 @@
 # Checks on every push to main and dev and every pull request: formatting, lints, tests, the
-# page's scripts, the pins (tokens, CodeMirror), the public-repository hygiene -- and the app
+# page's scripts, the pins (tokens, CodeMirror, xterm), the public-repository hygiene, the page
-# started for real, on X11 and on Wayland, to prove it gets as far as its window. The packages
+# itself under a headless browser (tests/ui) -- and the app started for real, on X11 and on
-# are built only by release.yml, from a tag.
+# Wayland, to prove it gets as far as its window. The packages are built only by release.yml,
+# from a tag.
 name: ci
 on:
   push:
@@ -40,11 +41,32 @@ jobs:
         run: cargo test --locked
       - name: The page's scripts parse, and none sets HTML
         run: scripts/check-ui.sh
-      - name: The design tokens and CodeMirror are the server's, as pinned
+      - name: The design tokens, CodeMirror and xterm are the server's, as pinned
         run: |
           scripts/fetch-tokens.sh
           scripts/fetch-vendor.sh
 
+  # The page (ui/) under a headless Chromium, against a stand-in for the app's bridge
+  # (tests/ui/mock-tauri.js) with the server's answers as a table: nothing is built and no server
+  # is started. Playwright is the library only, installed for the run and not a dependency of
+  # the repository (it has no package.json), with the Chromium it was made for.
+  ui:
+    runs-on: ubuntu-24.04
+    timeout-minutes: 20
+    steps:
+      - uses: actions/checkout@v5
+      - uses: actions/setup-node@v4
+        with:
+          node-version: 20
+      - name: Playwright and its Chromium
+        run: |
+          npm install --no-save --no-package-lock --prefix "$RUNNER_TEMP/pw" playwright-core@1.65.0
+          node "$RUNNER_TEMP/pw/node_modules/playwright-core/cli.js" install --with-deps chromium
+      - name: The page under a headless browser
+        env:
+          NODE_PATH: ${{ runner.temp }}/pw/node_modules
+        run: node tests/ui/run.mjs
+
   # The app started for real: it must still be running after 25 seconds (timeout's 124), and its
   # log must not say it stopped. Once under Xvfb (X11) and once under a headless Weston (Wayland,
   # GTK's Wayland backend forced), each with a session bus (the tray and the notifications want

CHANGELOG.md

@@ -4,7 +4,88 @@ All notable changes to LLeMbas Desktop. The format follows
 [Keep a Changelog](https://keepachangelog.com/en/1.1.0/), and the project uses
 [Semantic Versioning](https://semver.org/).
 
-## [Unreleased]
+## [1.0.0-beta.6] — 2026-11-10
+
+LLeMbas CLI 2.0.0-beta.6 inside, the look and words of LLeMbas 2.0.0-beta.8.
+
+### Fixed
+
+- **Approvals show again.** An agent asking to run a command, edit a file or carry out a plan
+  stopped with "Waiting for you to allow bash…" and no card to answer it: drawing the card's
+  buttons failed, and the space for it stayed empty. Stopping and asking again gave the same.
+  Now the card shows, with the command in the box to correct before it runs (it used to show the
+  words "Run a command" there instead), what the model says it is for, why it is asking, and which
+  subagent asks when one does. A card already on the screen is kept as it is while the reply goes
+  on, so a command being corrected or an option picked is no longer lost when the server sends
+  the list again; a card that cannot be drawn says why instead of leaving an empty space.
+- **Subagents after they are switched on.** The app read which models a new chat is offered
+  subagents on only when it signed in, so subagents switched on (or a model added) on the server
+  while the app stayed open in the tray never showed on a new chat until it was restarted. A new
+  chat now asks the server again. (An agent chat has no Subagents setting: its subagents run on
+  the computer, and the server offers none for it.)
+
+### Added
+
+- **Terminal.** An agent chat has a Terminal panel: a shell on the computer the chat works on, in
+  its folder, beside the chat (over it in a narrow window). A new agent chat offers it as soon as
+  the folder is chosen, before the first message -- and the chat that message makes carries on in
+  the same shell. An open agent chat opens its own from the header, the same shell the web
+  interface shows. **Send to chat** puts the selection, or the last command and its output, into
+  the message box to read before sending; the panel copies them too, and steps between commands
+  where the shell marks them. The terminal and the activity panel take turns beside the chat;
+  closing either keeps the shell. Offered where the computer allows a terminal (needs the server's
+  2.0.0-beta.8). The app opens the connection itself, with its sign-in: the page never holds it.
+  "Continue in a terminal" is now in the chat's menu only.
+- **What a reply is doing, the same way everywhere.** The line under a reply being written now
+  says it in the words every LLeMbas client uses (harness spec 3.4.0): queued, waiting for a
+  model and why, loading it, reading the prompt (with how far), thinking, writing (with its
+  speed), running a tool and what it runs, waiting for you, waiting for subagents, compacting,
+  retrying (counting down), stopped or failed -- each with how long it has lasted, kept moving
+  every second. From a server that sends no such line (before 2.0.0-beta.8), the app shows the
+  server's own words as before. In every language the server has.
+- **Subagents, the whole tree** (harness spec 3.4.0, needs the server's 2.0.0-beta.8). The
+  activity panel shows the main agent and every subagent under the one it reports to: its name and
+  id, its position, what it is doing right now, how full its context is, its state, model and
+  mode (with what it may ask for), its messages in the room and the last one, how many slots are
+  still open, and the ones that ended -- dimmed, with how they ended. Any subagent's conversation
+  opens read-only, live while it works and after it ended, and so do its findings; a working one
+  can be stopped (it and those under it). An agent chat's tree comes from its computer; while that
+  computer is away the panel shows the tree as it last was and says the rest is kept there.
+- **Subagents for agent chats.** A new agent chat offers the Subagents setting where the server
+  says this computer can run them, and an open agent chat where the server offers it for that
+  chat.
+- **A subagent's approval names it**: "Subagent parser (s1), reviews the parser, asks:".
+- `tests/ui/`: the page under a headless browser against a stand-in for the app's bridge, with
+  the server's answers as a table (`node tests/ui/run.mjs`, needs Playwright and a Chromium). CI
+  runs it on every push (the `ui` job), with Playwright and its Chromium installed for the run.
+
+## [1.0.0-beta.5] — 2026-11-09
+
+Carries LLeMbas CLI 2.0.0-beta.5 and the look of LLeMbas 2.0.0-beta.7, which the Subagents
+setting and the panel's compaction need.
+
+### Changed
+
+- **Subagents**, by that name everywhere. The chip under the message box says **Subagents** with the
+  least and the most agents a chat runs ("Subagents 1–4"), and opens one slider with two handles,
+  **Min** and **Max**, from 1 to the server's cap -- Min never passes Max, and both move with the
+  arrow keys as well. The numbers count the chat's own agent: **1 is the agent alone**, with no
+  subagents; Max 4 is the agent and up to 3 subagents. The chip shows where the
+  server offers subagents for the chat (for a new chat, for its model), and the most is the
+  server's cap. Where the chat's model serves one request at a time, it says that subagents need
+  another model, on another server, named in the chat. Subagents work on the chat's own model
+  unless you name another one in the chat -- there is no setting for a subagent's model. A
+  scheduled task's field is **Subagents** too. Needs LLeMbas 2.0.0-beta.7 for the server's word on
+  all of this; with an older one the app guesses as before.
+- **The activity panel, reshaped.** A button in the chat's header opens it at any time, and the
+  line above the message box shows only while something runs ("3 running · 1 queued · 2
+  finished"). An agent chat's **Environment** (its folder, and where it is) comes first. **Context**
+  says how far it is to the automatic compaction ("80.6k until auto-compact") beside a **Compact**
+  button that summarises the earlier turns now (greyed while there is nothing to summarise). **Background tasks** lists subagents, a team's
+  members and background commands under **Running**, each with how long it has run and a stop
+  square, unfolding to what it is, its latest step, what it is writing and its conversation; the
+  tasks under one fold beneath it ("Tasks: 2"), and the finished ones fold away under **Finished**.
+  The team's **Room** and the **Plan** have sections of their own.
 
 ## [1.0.0-beta.4] — 2026-11-09
 

lembas-cli.version

@@ -1 +1 @@
-2.0.0-beta.3
+2.0.0-beta.6

scripts/fetch-fixtures.sh

@@ -0,0 +1,27 @@
+#!/usr/bin/env bash
+# Copy the server's /api/v2 contract fixtures that tests/ui reads into tests/ui/fixtures/:
+#
+#   scripts/fetch-fixtures.sh <server-ref>     (a tag such as v2.0.0-beta.8, or a branch)
+#
+# The server is a git checkout, LLeMbas beside this repository by default (LEMBAS_SERVER names
+# another); the files are read from the ref, not from its working tree. SOURCE records the ref and
+# the commit, so the tests say which server they hold the app to. CI has no server checkout: the
+# copies are committed.
+set -euo pipefail
+cd "$(dirname "$0")/.."
+
+SERVER="${LEMBAS_SERVER:-../LLeMbas}"
+FROM="tests/fixtures/api_v2"
+OUT="tests/ui/fixtures"
+FILES="chat_agent.json event_ask.json event_status.json event_status_idle.json event_status_person.json
+event_status_stopped.json event_tree.json models_agent.json subagent_findings.json subagent_transcript.json tree.json"
+
+[ $# -eq 1 ] || { echo "usage: scripts/fetch-fixtures.sh <server-ref>" >&2; exit 2; }
+commit="$(git -C "$SERVER" rev-parse --verify --quiet "$1^{commit}")" \
+  || { echo "fetch-fixtures: $1 is not a commit in $SERVER" >&2; exit 1; }
+mkdir -p "$OUT"
+for f in $FILES; do
+  git -C "$SERVER" show "$commit:$FROM/$f" > "$OUT/$f" || { echo "fetch-fixtures: $f is not at $1" >&2; exit 1; }
+done
+printf 'LLeMbas %s (%s) %s\n' "$1" "$commit" "$FROM" > "$OUT/SOURCE"
+echo "fetch-fixtures: $(echo $FILES | wc -w) files from LLeMbas $1 (${commit:0:12}) -> $OUT"

scripts/fetch-vendor.sh

@@ -1,6 +1,8 @@
 #!/usr/bin/env bash
-# Put the server's vendored CodeMirror build where the app's own pages read it from:
+# Put the server's vendored CodeMirror build and xterm where the app's own pages read them from:
 #   ui/vendor/*.mjs   (copies of the server's static/vendor/cm-*.mjs and lezer-*.mjs)
+#   ui/vendor/xterm.js, xterm.css, xterm-addon-fit.js   (the terminal panel's, copied unchanged:
+#                     classic scripts the panel loads when it is first opened)
 #
 #   scripts/fetch-vendor.sh <server-ref>      copy the modules at that ref (a tag such as
 #                                             v2.0.0-beta.1, or a branch) and write vendor.lock.json
@@ -47,8 +49,8 @@ if [ $# -eq 0 ]; then
       echo "fetch-vendor: $OUT/$name does not match $LOCK (server $ref)" >&2
       bad=1
     fi
-  done < <(sed -n -E 's/^ *"([a-z0-9-]+\.mjs)": *\{"server": *"[0-9a-f]{64}", *"copy": *"([0-9a-f]{64})"\}.*/\1\t\2/p' "$LOCK")
+  done < <(sed -n -E 's/^ *"([a-z0-9-]+\.(mjs|js|css))": *\{"server": *"[0-9a-f]{64}", *"copy": *"([0-9a-f]{64})"\}.*/\1\t\3/p' "$LOCK")
-  for f in "$OUT"/*.mjs; do
+  for f in "$OUT"/*; do
     grep -q "\"$(basename "$f")\"" "$LOCK" || { echo "fetch-vendor: $f is not in $LOCK" >&2; bad=1; }
   done
   [ "$count" -gt 0 ] || { echo "fetch-vendor: $LOCK lists no file" >&2; exit 1; }
@@ -87,6 +89,16 @@ for name in $names; do
   fi
   entries+=("$(printf '    "%s": {"server": "%s", "copy": "%s"}' "$name" "$have" "$(sha "$work/out/$name")")")
 done
+# xterm and its fit addon, for the terminal panel: copied as they are, each checked against the
+# server's pin like the modules.
+for name in xterm.js xterm.css xterm-addon-fit.js; do
+  git -C "$SERVER" show "$commit:$SOURCE/$name" > "$work/out/$name" \
+    || { echo "fetch-vendor: $name is not in $SOURCE at $REF" >&2; exit 1; }
+  pinned="$(python3 -c 'import json,sys; print(json.load(open(sys.argv[1])).get(sys.argv[2], {}).get("sha256", ""))' "$work/server-lock.json" "$name")"
+  have="$(sha "$work/out/$name")"
+  [ "$pinned" = "$have" ] || { echo "fetch-vendor: $name does not match the server's pin ($have, pinned $pinned)" >&2; exit 1; }
+  entries+=("$(printf '    "%s": {"server": "%s", "copy": "%s"}' "$name" "$have" "$have")")
+done
 # The polyfill stub, from the server's own scripts (it is not a download, so it has no pin there).
 stub="src/lembas/web/static/js/node-process.mjs"
 git -C "$SERVER" show "$commit:$stub" > "$work/out/node-process.mjs" \
@@ -101,7 +113,7 @@ done
 
 rm -rf "$OUT"
 mkdir -p "$OUT"
-cp "$work"/out/*.mjs "$OUT/"
+cp "$work"/out/*.mjs "$work"/out/xterm.js "$work"/out/xterm.css "$work"/out/xterm-addon-fit.js "$OUT/"
 {
   printf '{\n  "tag": "%s",\n  "commit": "%s",\n  "files": {\n' "$REF" "$commit"
   last=$((${#entries[@]} - 1))

scripts/gen-i18n.py

@@ -61,6 +61,8 @@ KEYS = (
     "The agent asks to run %(tool)s:",
     "The agent asks:",
     "Command",
+    "%(job)s asks:",
+    "It says: %(purpose)s",
     "Recommended",
     "Or type an answer",
     "Your answer",
@@ -194,10 +196,8 @@ KEYS = (
     "off",
     "default",
     "Think before answering",
-    "Fewest",
+    "Min",
-    "Most",
+    "Max",
-    "Agents",
-    "Its own agent included. Above one, it can add members that do the work while it leads them.",
     "Continue in a terminal…",
     "Continue in a terminal",
     "Run this in a terminal: the agent's session goes on there, in this chat's folder, and the chat keeps it.",
@@ -207,8 +207,6 @@ KEYS = (
     "Back",
     "Open the activity panel",
     "Close the activity panel",
-    "1 agent working",
-    "%(n)s agents working",
     "%(n)s queued",
     "%(n)s waiting",
     "%(n)s finished",
@@ -240,9 +238,7 @@ KEYS = (
     "Estimated: no reply has been counted yet.",
     "%(n)s tools offered.",
     "No files written in this chat yet.",
-    "Nothing runs in the background. Helpers, commands and a team's members show here while they work.",
     "Nothing running.",
-    "Team: %(size)s of %(max)s agents",
     "Show %(n)s more",
     "Room",
     "report",
@@ -323,8 +319,6 @@ KEYS = (
     "A busy model is waited for, then the next free one is used.",
     "A result, but not a run with nothing to report.",
     "A task without a model of its own runs on the first free one of these, in this order. None ticked: all of them.",
-    "A team of up to %(n)s",
-    "A team works on it together, its own agent included.",
     "Choose a model to choose its effort.",
     "Due %(when)s",
     "Each run sees the ones before it.",
@@ -333,7 +327,6 @@ KEYS = (
     "Edit %(title)s",
     "Every result, even with nothing to report.",
     "Models for scheduled tasks",
-    "No team",
     "Only in its history.",
     "Open the chat",
     "Result",
@@ -358,6 +351,100 @@ KEYS = (
     "Open its chat",
     "The model's own",
     "run by hand",
+    # activity.js, chat.js, models.js, scheduled.js: subagents, the panel, Compact
+    "%(n)s running",
+    "Tasks: %(n)s",
+    "%(tokens)s until auto-compact",
+    "Automatic compaction is off.",
+    "Compacts before the next reply.",
+    "Compacts by itself once the window is known.",
+    "Compact",
+    "Compacting…",
+    "Summarise the earlier turns now and send the summary in their place.",
+    "This server cannot compact a chat from the app yet; the web interface can.",
+    "Context and background tasks",
+    "Environment",
+    "Finished %(n)s",
+    "Background command",
+    "Subagent",
+    "Team member",
+    "Subagents",
+    "The numbers count this chat's own agent: 1 is the agent alone, with no subagents; Max 4 is the agent and up to 3 subagents. A subagent runs on the chat's model, or on another only when you name it in the chat.",
+    "This model serves one request at a time: subagents need another model, on another server, named in the chat.",
+    "None",
+    "Up to %(n)s at once",
+    "Nothing runs in the background. Subagents, a team's members and background commands show here while they work.",
+    "Team: %(size)s of at most %(max)s subagents",
+    "How many work on it at once, the run's own agent among them, on the same model.",
+    # terminal.js
+    "Terminal",
+    "Close terminal",
+    "Send to chat",
+    "Connecting…",
+    "Connected.",
+    "Connected. This shell is also open in another tab, and they share a size.",
+    "Disconnected. Close and reopen to reconnect.",
+    "This terminal closed.",
+    "Reconnecting…",
+    "The terminal could not be loaded.",
+    "Nothing has been run in this shell yet.",
+    "Nothing to send: select some output first.",
+    "Copied the last of the screen, as it appeared. This shell does not mark where commands begin.",
+    "Copied.",
+    "The previous command",
+    "The previous command (Ctrl+Shift+↑)",
+    "The next command",
+    "The next command (Ctrl+Shift+↓)",
+    "Copy the last command and its output",
+    "Put the last command and its output into the message box",
+    "Sign in again: the server no longer accepts this app's sign-in.",
+    "The terminal needs a newer LLeMbas server. Ask its administrator to update it.",
+    # progress.js (harness spec 3.4.0: loop.json progress.texts, keyed as the spec writes them)
+    "Queued",
+    "Queued: {{ahead}} ahead",
+    "Waiting for {{model}}: {{reason}}",
+    "Loading {{model}}",
+    "Reading the prompt: {{done}} / {{total}} tokens ({{percent}}%)",
+    "Reading the prompt",
+    "Writing: {{rate}} tok/s",
+    "Writing",
+    "Running {{tool}}: {{what}}",
+    "Running {{tool}}",
+    "Waiting for you: {{what}}",
+    "Waiting for subagents: {{running}} running",
+    "Compacting the conversation",
+    "Retrying in {{in}} s: {{reason}}",
+    "Stopped",
+    "Stopped: {{reason}}",
+    "Failed: {{reason}}",
+    "it serves one request at a time and is answering another",
+    "all its slots are in use",
+    "its server holds another model now",
+    "allow {{tool}}?",
+    "an approval",
+    "a question",
+    "the plan",
+    "by you",
+    "step limit",
+    "budget",
+    # tree.js, activity.js (the subagent tree), chat.js (a subagent's approval)
+    "waiting for a slot",
+    "finished",
+    "idle",
+    "paused: context full",
+    "out of context",
+    "the main agent",
+    "%(size)s of %(max)s",
+    "Open slots: %(n)s",
+    "They run on the computer this chat runs on.",
+    "Ended",
+    "Context used",
+    "messages: %(n)s",
+    "findings: %(n)s",
+    "Findings",
+    "Stop %(name)s and the subagents under it?",
+    "This device is not linked right now.",
+    "Subagent %(name)s (%(id)s), %(position)s, asks:",
 )
 
 # The app's own sentences. The mode names (Manual, Edit, Auto, Plan) stay English, as the
@@ -365,6 +452,34 @@ KEYS = (
 # the device, as the server's catalogue has it.
 OWN: dict[str, dict[str, str]] = {
     "cs": {
+        "Sign in again: the server no longer accepts this app's sign-in.": "Přihlaste se znovu: server už nepřijímá přihlášení této aplikace.",
+        "The terminal needs a newer LLeMbas server. Ask its administrator to update it.": "Terminál potřebuje novější server LLeMbas. Požádejte jeho správce o aktualizaci.",
+        "This model serves one request at a time: subagents need another model, on another server, named in the chat.": "Tento model vyřizuje najednou jeden požadavek: podagenti potřebují jiný model na jiném serveru, pojmenovaný v konverzaci.",
+        "The numbers count this chat's own agent: 1 is the agent alone, with no subagents; Max 4 is the agent and up to 3 subagents. A subagent runs on the chat's model, or on another only when you name it in the chat.": "Čísla počítají i vlastního agenta konverzace: 1 je agent sám, bez podagentů; Max 4 je agent a až 3 podagenti. Podagent běží na modelu konverzace, na jiném jen tehdy, když ho v konverzaci pojmenujete.",
+        "Min": "Min",
+        "Max": "Max",
+        "Tasks: %(n)s": "Úlohy: %(n)s",
+        "%(n)s running": "Běží: %(n)s",
+        "%(tokens)s until auto-compact": "%(tokens)s do automatického shrnutí",
+        "Automatic compaction is off.": "Automatické shrnutí je vypnuté.",
+        "Compacts before the next reply.": "Shrne se před další odpovědí.",
+        "Compacts by itself once the window is known.": "Shrne se samo, jakmile bude známé okno.",
+        "Compact": "Shrnout",
+        "Compacting…": "Shrnuje se…",
+        "Summarise the earlier turns now and send the summary in their place.": "Shrnout dřívější kola hned a posílat místo nich shrnutí.",
+        "This server cannot compact a chat from the app yet; the web interface can.": "Tento server zatím neumí shrnout konverzaci z aplikace; webové rozhraní to umí.",
+        "Context and background tasks": "Kontext a úlohy na pozadí",
+        "Environment": "Prostředí",
+        "Finished %(n)s": "Dokončeno: %(n)s",
+        "Background command": "Příkaz na pozadí",
+        "Subagent": "Podagent",
+        "Team member": "Člen týmu",
+        "Subagents": "Podagenti",
+        "None": "Žádní",
+        "Up to %(n)s at once": "Nejvýše %(n)s současně",
+        "Nothing runs in the background. Subagents, a team's members and background commands show here while they work.": "Na pozadí nic neběží. Podagenti, členové týmu a příkazy na pozadí se tu zobrazí, dokud pracují.",
+        "Team: %(size)s of at most %(max)s subagents": "Tým: %(size)s z nejvýše %(max)s podagentů",
+        "How many work on it at once, the run's own agent among them, on the same model.": "Kolik jich na ní pracuje současně, včetně vlastního agenta běhu, na stejném modelu.",
         "Menu": "Nabídka",
         "Minimise": "Minimalizovat",
         "Maximise": "Maximalizovat",
@@ -532,8 +647,6 @@ OWN: dict[str, dict[str, str]] = {
         "A busy model is waited for, then the next free one is used.": "Na zaneprázdněný model se čeká, pak se použije další volný.",
         "A result, but not a run with nothing to report.": "Výsledek, ale ne spuštění, které nemá co hlásit.",
         "A task without a model of its own runs on the first free one of these, in this order. None ticked: all of them.": "Úloha bez vlastního modelu běží na prvním volném z těchto, v tomto pořadí. Žádný nezaškrtnutý: všechny.",
-        "A team of up to %(n)s": "Tým: nejvýše %(n)s",
-        "A team works on it together, its own agent included.": "Pracuje na ní tým společně, včetně jejího vlastního agenta.",
         "Choose a model to choose its effort.": "Vyberte model, abyste mohli zvolit jeho náročnost.",
         "Due %(when)s": "Naplánováno na %(when)s",
         "Each run sees the ones before it.": "Každé spuštění vidí ta předchozí.",
@@ -542,7 +655,6 @@ OWN: dict[str, dict[str, str]] = {
         "Edit %(title)s": "Upravit %(title)s",
         "Every result, even with nothing to report.": "Každý výsledek, i když není co hlásit.",
         "Models for scheduled tasks": "Modely pro naplánované úlohy",
-        "No team": "Bez týmu",
         "Only in its history.": "Jen v její historii.",
         "Open the chat": "Otevřít konverzaci",
         "Result": "Výsledek",
@@ -563,6 +675,34 @@ OWN: dict[str, dict[str, str]] = {
         "No day ticked: every day.": "Žádný den nezaškrtnutý: každý den.",
     },
     "de": {
+        "Sign in again: the server no longer accepts this app's sign-in.": "Melden Sie sich erneut an: Der Server nimmt die Anmeldung dieser App nicht mehr an.",
+        "The terminal needs a newer LLeMbas server. Ask its administrator to update it.": "Das Terminal braucht einen neueren LLeMbas-Server. Bitten Sie seinen Administrator, ihn zu aktualisieren.",
+        "This model serves one request at a time: subagents need another model, on another server, named in the chat.": "Dieses Modell bedient eine Anfrage zur Zeit: Subagenten brauchen ein anderes Modell auf einem anderen Server, im Chat genannt.",
+        "The numbers count this chat's own agent: 1 is the agent alone, with no subagents; Max 4 is the agent and up to 3 subagents. A subagent runs on the chat's model, or on another only when you name it in the chat.": "Die Zahlen zählen den eigenen Agenten des Chats mit: 1 ist der Agent allein, ohne Subagenten; Max 4 ist der Agent und bis zu 3 Subagenten. Ein Subagent läuft auf dem Modell des Chats, auf einem anderen nur, wenn Sie es im Chat nennen.",
+        "Min": "Min",
+        "Max": "Max",
+        "Tasks: %(n)s": "Aufgaben: %(n)s",
+        "%(n)s running": "%(n)s laufen",
+        "%(tokens)s until auto-compact": "%(tokens)s bis zur automatischen Zusammenfassung",
+        "Automatic compaction is off.": "Die automatische Zusammenfassung ist aus.",
+        "Compacts before the next reply.": "Wird vor der nächsten Antwort zusammengefasst.",
+        "Compacts by itself once the window is known.": "Wird von selbst zusammengefasst, sobald das Fenster bekannt ist.",
+        "Compact": "Zusammenfassen",
+        "Compacting…": "Wird zusammengefasst…",
+        "Summarise the earlier turns now and send the summary in their place.": "Die früheren Runden jetzt zusammenfassen und stattdessen die Zusammenfassung senden.",
+        "This server cannot compact a chat from the app yet; the web interface can.": "Dieser Server kann einen Chat noch nicht aus der App zusammenfassen; die Weboberfläche kann es.",
+        "Context and background tasks": "Kontext und Hintergrundaufgaben",
+        "Environment": "Umgebung",
+        "Finished %(n)s": "Fertig: %(n)s",
+        "Background command": "Hintergrundbefehl",
+        "Subagent": "Subagent",
+        "Team member": "Teammitglied",
+        "Subagents": "Subagenten",
+        "None": "Keine",
+        "Up to %(n)s at once": "Bis zu %(n)s gleichzeitig",
+        "Nothing runs in the background. Subagents, a team's members and background commands show here while they work.": "Im Hintergrund läuft nichts. Subagenten, die Mitglieder eines Teams und Hintergrundbefehle erscheinen hier, solange sie arbeiten.",
+        "Team: %(size)s of at most %(max)s subagents": "Team: %(size)s von höchstens %(max)s Subagenten",
+        "How many work on it at once, the run's own agent among them, on the same model.": "Wie viele gleichzeitig daran arbeiten, der eigene Agent des Laufs mitgezählt, mit demselben Modell.",
         "Menu": "Menü",
         "Minimise": "Minimieren",
         "Maximise": "Maximieren",
@@ -730,8 +870,6 @@ OWN: dict[str, dict[str, str]] = {
         "A busy model is waited for, then the next free one is used.": "Auf ein belegtes Modell wird gewartet, dann wird das nächste freie genommen.",
         "A result, but not a run with nothing to report.": "Ein Ergebnis, aber kein Lauf ohne Neuigkeiten.",
         "A task without a model of its own runs on the first free one of these, in this order. None ticked: all of them.": "Eine Aufgabe ohne eigenes Modell läuft auf dem ersten freien davon, in dieser Reihenfolge. Keines angehakt: alle.",
-        "A team of up to %(n)s": "Ein Team mit bis zu %(n)s",
-        "A team works on it together, its own agent included.": "Ein Team arbeitet gemeinsam daran, ihr eigener Agent eingeschlossen.",
         "Choose a model to choose its effort.": "Wählen Sie ein Modell, um seinen Aufwand zu wählen.",
         "Due %(when)s": "Fällig %(when)s",
         "Each run sees the ones before it.": "Jeder Lauf sieht die vorherigen.",
@@ -740,7 +878,6 @@ OWN: dict[str, dict[str, str]] = {
         "Edit %(title)s": "%(title)s bearbeiten",
         "Every result, even with nothing to report.": "Jedes Ergebnis, auch ohne Neuigkeiten.",
         "Models for scheduled tasks": "Modelle für geplante Aufgaben",
-        "No team": "Kein Team",
         "Only in its history.": "Nur in ihrem Verlauf.",
         "Open the chat": "Chat öffnen",
         "Result": "Ergebnis",
@@ -761,6 +898,34 @@ OWN: dict[str, dict[str, str]] = {
         "No day ticked: every day.": "Kein Tag angehakt: jeden Tag.",
     },
     "es": {
+        "Sign in again: the server no longer accepts this app's sign-in.": "Vuelva a iniciar sesión: el servidor ya no acepta el inicio de sesión de esta aplicación.",
+        "The terminal needs a newer LLeMbas server. Ask its administrator to update it.": "El terminal necesita un servidor LLeMbas más reciente. Pida a su administrador que lo actualice.",
+        "This model serves one request at a time: subagents need another model, on another server, named in the chat.": "Este modelo atiende una petición a la vez: los subagentes necesitan otro modelo, en otro servidor, nombrado en el chat.",
+        "The numbers count this chat's own agent: 1 is the agent alone, with no subagents; Max 4 is the agent and up to 3 subagents. A subagent runs on the chat's model, or on another only when you name it in the chat.": "Los números cuentan el agente propio del chat: 1 es el agente solo, sin subagentes; Máx. 4 es el agente y hasta 3 subagentes. Un subagente usa el modelo del chat, u otro solo si lo nombra en el chat.",
+        "Min": "Mín.",
+        "Max": "Máx.",
+        "Tasks: %(n)s": "Tareas: %(n)s",
+        "%(n)s running": "%(n)s en marcha",
+        "%(tokens)s until auto-compact": "%(tokens)s hasta la compactación automática",
+        "Automatic compaction is off.": "La compactación automática está desactivada.",
+        "Compacts before the next reply.": "Se compacta antes de la próxima respuesta.",
+        "Compacts by itself once the window is known.": "Se compacta sola en cuanto se conozca la ventana.",
+        "Compact": "Compactar",
+        "Compacting…": "Compactando…",
+        "Summarise the earlier turns now and send the summary in their place.": "Resumir ahora los turnos anteriores y enviar el resumen en su lugar.",
+        "This server cannot compact a chat from the app yet; the web interface can.": "Este servidor aún no puede compactar un chat desde la aplicación; la interfaz web sí puede.",
+        "Context and background tasks": "Contexto y tareas en segundo plano",
+        "Environment": "Entorno",
+        "Finished %(n)s": "Terminadas: %(n)s",
+        "Background command": "Comando en segundo plano",
+        "Subagent": "Subagente",
+        "Team member": "Miembro del equipo",
+        "Subagents": "Subagentes",
+        "None": "Ninguno",
+        "Up to %(n)s at once": "Hasta %(n)s a la vez",
+        "Nothing runs in the background. Subagents, a team's members and background commands show here while they work.": "No hay nada en segundo plano. Los subagentes, los miembros de un equipo y los comandos en segundo plano aparecen aquí mientras trabajan.",
+        "Team: %(size)s of at most %(max)s subagents": "Equipo: %(size)s de un máximo de %(max)s subagentes",
+        "How many work on it at once, the run's own agent among them, on the same model.": "Cuántos trabajan en ella a la vez, contando el propio agente de la ejecución, con el mismo modelo.",
         "Menu": "Menú",
         "Minimise": "Minimizar",
         "Maximise": "Maximizar",
@@ -928,8 +1093,6 @@ OWN: dict[str, dict[str, str]] = {
         "A busy model is waited for, then the next free one is used.": "Se espera a un modelo ocupado y luego se usa el siguiente libre.",
         "A result, but not a run with nothing to report.": "Un resultado, pero no una ejecución sin nada que informar.",
         "A task without a model of its own runs on the first free one of these, in this order. None ticked: all of them.": "Una tarea sin modelo propio se ejecuta en el primero libre de estos, en este orden. Ninguno marcado: todos.",
-        "A team of up to %(n)s": "Un equipo de hasta %(n)s",
-        "A team works on it together, its own agent included.": "Un equipo trabaja en ella, incluido su propio agente.",
         "Choose a model to choose its effort.": "Elija un modelo para elegir su esfuerzo.",
         "Due %(when)s": "Prevista %(when)s",
         "Each run sees the ones before it.": "Cada ejecución ve las anteriores.",
@@ -938,7 +1101,6 @@ OWN: dict[str, dict[str, str]] = {
         "Edit %(title)s": "Editar %(title)s",
         "Every result, even with nothing to report.": "Cada resultado, incluso sin nada que informar.",
         "Models for scheduled tasks": "Modelos para tareas programadas",
-        "No team": "Sin equipo",
         "Only in its history.": "Solo en su historial.",
         "Open the chat": "Abrir el chat",
         "Result": "Resultado",
@@ -959,6 +1121,34 @@ OWN: dict[str, dict[str, str]] = {
         "No day ticked: every day.": "Ningún día marcado: todos los días.",
     },
     "fr": {
+        "Sign in again: the server no longer accepts this app's sign-in.": "Reconnectez-vous : le serveur n’accepte plus la connexion de cette application.",
+        "The terminal needs a newer LLeMbas server. Ask its administrator to update it.": "Le terminal nécessite un serveur LLeMbas plus récent. Demandez à son administrateur de le mettre à jour.",
+        "This model serves one request at a time: subagents need another model, on another server, named in the chat.": "Ce modèle traite une requête à la fois : les sous-agents ont besoin d’un autre modèle, sur un autre serveur, nommé dans la conversation.",
+        "The numbers count this chat's own agent: 1 is the agent alone, with no subagents; Max 4 is the agent and up to 3 subagents. A subagent runs on the chat's model, or on another only when you name it in the chat.": "Les nombres comptent l’agent de la conversation : 1, c’est l’agent seul, sans sous-agents ; Max 4, c’est l’agent et jusqu’à 3 sous-agents. Un sous-agent utilise le modèle de la conversation, ou un autre seulement si vous le nommez dans la conversation.",
+        "Min": "Min",
+        "Max": "Max",
+        "Tasks: %(n)s": "Tâches : %(n)s",
+        "%(n)s running": "%(n)s en cours",
+        "%(tokens)s until auto-compact": "%(tokens)s avant le compactage automatique",
+        "Automatic compaction is off.": "Le compactage automatique est désactivé.",
+        "Compacts before the next reply.": "Compacté avant la prochaine réponse.",
+        "Compacts by itself once the window is known.": "Compacté de lui-même dès que la fenêtre est connue.",
+        "Compact": "Compacter",
+        "Compacting…": "Compactage…",
+        "Summarise the earlier turns now and send the summary in their place.": "Résumer maintenant les tours précédents et envoyer le résumé à leur place.",
+        "This server cannot compact a chat from the app yet; the web interface can.": "Ce serveur ne peut pas encore compacter une conversation depuis l’application ; l’interface web le peut.",
+        "Context and background tasks": "Contexte et tâches en arrière-plan",
+        "Environment": "Environnement",
+        "Finished %(n)s": "Terminées : %(n)s",
+        "Background command": "Commande en arrière-plan",
+        "Subagent": "Sous-agent",
+        "Team member": "Membre de l’équipe",
+        "Subagents": "Sous-agents",
+        "None": "Aucun",
+        "Up to %(n)s at once": "Jusqu’à %(n)s à la fois",
+        "Nothing runs in the background. Subagents, a team's members and background commands show here while they work.": "Rien ne tourne en arrière-plan. Les sous-agents, les membres d’une équipe et les commandes en arrière-plan s’affichent ici pendant qu’ils travaillent.",
+        "Team: %(size)s of at most %(max)s subagents": "Équipe : %(size)s sur %(max)s sous-agents au plus",
+        "How many work on it at once, the run's own agent among them, on the same model.": "Combien y travaillent à la fois, l’agent de l’exécution compris, avec le même modèle.",
         "Menu": "Menu",
         "Minimise": "Réduire",
         "Maximise": "Agrandir",
@@ -1126,8 +1316,6 @@ OWN: dict[str, dict[str, str]] = {
         "A busy model is waited for, then the next free one is used.": "Un modèle occupé est attendu, puis le suivant libre est utilisé.",
         "A result, but not a run with nothing to report.": "Un résultat, mais pas une exécution sans rien à signaler.",
         "A task without a model of its own runs on the first free one of these, in this order. None ticked: all of them.": "Une tâche sans modèle à elle tourne sur le premier libre de ceux-ci, dans cet ordre. Aucun coché : tous.",
-        "A team of up to %(n)s": "Une équipe jusqu'à %(n)s",
-        "A team works on it together, its own agent included.": "Une équipe y travaille ensemble, son propre agent compris.",
         "Choose a model to choose its effort.": "Choisissez un modèle pour choisir son effort.",
         "Due %(when)s": "Prévue %(when)s",
         "Each run sees the ones before it.": "Chaque exécution voit les précédentes.",
@@ -1136,7 +1324,6 @@ OWN: dict[str, dict[str, str]] = {
         "Edit %(title)s": "Modifier %(title)s",
         "Every result, even with nothing to report.": "Chaque résultat, même sans rien à signaler.",
         "Models for scheduled tasks": "Modèles pour les tâches planifiées",
-        "No team": "Pas d'équipe",
         "Only in its history.": "Seulement dans son historique.",
         "Open the chat": "Ouvrir la conversation",
         "Result": "Résultat",
@@ -1157,6 +1344,34 @@ OWN: dict[str, dict[str, str]] = {
         "No day ticked: every day.": "Aucun jour coché : tous les jours.",
     },
     "hu": {
+        "Sign in again: the server no longer accepts this app's sign-in.": "Jelentkezzen be újra: a szerver már nem fogadja el ennek az alkalmazásnak a bejelentkezését.",
+        "The terminal needs a newer LLeMbas server. Ask its administrator to update it.": "A terminálhoz újabb LLeMbas-szerver kell. Kérje meg a rendszergazdáját, hogy frissítse.",
+        "This model serves one request at a time: subagents need another model, on another server, named in the chat.": "Ez a modell egyszerre egy kérést szolgál ki: az alügynököknek másik modell kell, másik szerveren, a beszélgetésben megnevezve.",
+        "The numbers count this chat's own agent: 1 is the agent alone, with no subagents; Max 4 is the agent and up to 3 subagents. A subagent runs on the chat's model, or on another only when you name it in the chat.": "A számokba a beszélgetés saját ügynöke is beleszámít: az 1 maga az ügynök, alügynökök nélkül; a Max. 4 az ügynök és legfeljebb 3 alügynök. Az alügynök a beszélgetés modelljén fut, másikon csak akkor, ha megnevezi a beszélgetésben.",
+        "Min": "Min.",
+        "Max": "Max.",
+        "Tasks: %(n)s": "Feladatok: %(n)s",
+        "%(n)s running": "%(n)s fut",
+        "%(tokens)s until auto-compact": "%(tokens)s az automatikus tömörítésig",
+        "Automatic compaction is off.": "Az automatikus tömörítés ki van kapcsolva.",
+        "Compacts before the next reply.": "A következő válasz előtt tömörül.",
+        "Compacts by itself once the window is known.": "Magától tömörül, amint ismert az ablak.",
+        "Compact": "Tömörítés",
+        "Compacting…": "Tömörítés…",
+        "Summarise the earlier turns now and send the summary in their place.": "A korábbi köröket most összefoglalja, és helyettük az összefoglalót küldi.",
+        "This server cannot compact a chat from the app yet; the web interface can.": "Ez a szerver még nem tud beszélgetést tömöríteni az alkalmazásból; a webes felület tud.",
+        "Context and background tasks": "Kontextus és háttérfeladatok",
+        "Environment": "Környezet",
+        "Finished %(n)s": "Kész: %(n)s",
+        "Background command": "Háttérparancs",
+        "Subagent": "Alügynök",
+        "Team member": "Csapattag",
+        "Subagents": "Alügynökök",
+        "None": "Nincs",
+        "Up to %(n)s at once": "Legfeljebb %(n)s egyszerre",
+        "Nothing runs in the background. Subagents, a team's members and background commands show here while they work.": "A háttérben semmi sem fut. Az alügynökök, egy csapat tagjai és a háttérparancsok itt jelennek meg, amíg dolgoznak.",
+        "Team: %(size)s of at most %(max)s subagents": "Csapat: %(size)s a legfeljebb %(max)s alügynökből",
+        "How many work on it at once, the run's own agent among them, on the same model.": "Hányan dolgoznak rajta egyszerre, a futás saját ügynökét is beleértve, ugyanazon a modellen.",
         "Menu": "Menü",
         "Minimise": "Kis méret",
         "Maximise": "Teljes méret",
@@ -1324,8 +1539,6 @@ OWN: dict[str, dict[str, str]] = {
         "A busy model is waited for, then the next free one is used.": "A foglalt modellre vár, majd a következő szabadot használja.",
         "A result, but not a run with nothing to report.": "Eredmény, de nem olyan futás, amelynek nincs mit jelentenie.",
         "A task without a model of its own runs on the first free one of these, in this order. None ticked: all of them.": "A saját modell nélküli feladat ezek közül az első szabadon fut, ebben a sorrendben. Ha egy sincs bejelölve: mindegyiken.",
-        "A team of up to %(n)s": "Legfeljebb %(n)s fős csapat",
-        "A team works on it together, its own agent included.": "Egy csapat dolgozik rajta együtt, a saját ügynökét is beleértve.",
         "Choose a model to choose its effort.": "Válasszon modellt az erőfeszítés megadásához.",
         "Due %(when)s": "Esedékes: %(when)s",
         "Each run sees the ones before it.": "Minden futás látja az előzőeket.",
@@ -1334,7 +1547,6 @@ OWN: dict[str, dict[str, str]] = {
         "Edit %(title)s": "%(title)s szerkesztése",
         "Every result, even with nothing to report.": "Minden eredmény, akkor is, ha nincs mit jelenteni.",
         "Models for scheduled tasks": "Modellek az ütemezett feladatokhoz",
-        "No team": "Nincs csapat",
         "Only in its history.": "Csak az előzményeiben.",
         "Open the chat": "A csevegés megnyitása",
         "Result": "Eredmény",
@@ -1355,6 +1567,34 @@ OWN: dict[str, dict[str, str]] = {
         "No day ticked: every day.": "Ha egy nap sincs bejelölve: minden nap.",
     },
     "it": {
+        "Sign in again: the server no longer accepts this app's sign-in.": "Accedi di nuovo: il server non accetta più l’accesso di questa app.",
+        "The terminal needs a newer LLeMbas server. Ask its administrator to update it.": "Il terminale richiede un server LLeMbas più recente. Chiedi al suo amministratore di aggiornarlo.",
+        "This model serves one request at a time: subagents need another model, on another server, named in the chat.": "Questo modello serve una richiesta alla volta: i sottoagenti hanno bisogno di un altro modello, su un altro server, nominato nella chat.",
+        "The numbers count this chat's own agent: 1 is the agent alone, with no subagents; Max 4 is the agent and up to 3 subagents. A subagent runs on the chat's model, or on another only when you name it in the chat.": "I numeri contano anche l’agente della chat: 1 è l’agente da solo, senza sottoagenti; Max 4 è l’agente più fino a 3 sottoagenti. Un sottoagente usa il modello della chat, o un altro solo se lo nomini nella chat.",
+        "Min": "Min",
+        "Max": "Max",
+        "Tasks: %(n)s": "Attività: %(n)s",
+        "%(n)s running": "%(n)s in corso",
+        "%(tokens)s until auto-compact": "%(tokens)s alla compattazione automatica",
+        "Automatic compaction is off.": "La compattazione automatica è disattivata.",
+        "Compacts before the next reply.": "Si compatta prima della prossima risposta.",
+        "Compacts by itself once the window is known.": "Si compatta da sola appena la finestra è nota.",
+        "Compact": "Compatta",
+        "Compacting…": "Compattazione…",
+        "Summarise the earlier turns now and send the summary in their place.": "Riassumi ora i turni precedenti e invia il riassunto al loro posto.",
+        "This server cannot compact a chat from the app yet; the web interface can.": "Questo server non può ancora compattare una chat dall’app; l’interfaccia web sì.",
+        "Context and background tasks": "Contesto e attività in background",
+        "Environment": "Ambiente",
+        "Finished %(n)s": "Terminate: %(n)s",
+        "Background command": "Comando in background",
+        "Subagent": "Sottoagente",
+        "Team member": "Membro della squadra",
+        "Subagents": "Sottoagenti",
+        "None": "Nessuno",
+        "Up to %(n)s at once": "Fino a %(n)s alla volta",
+        "Nothing runs in the background. Subagents, a team's members and background commands show here while they work.": "Nulla è in esecuzione in background. Sottoagenti, membri di una squadra e comandi in background compaiono qui mentre lavorano.",
+        "Team: %(size)s of at most %(max)s subagents": "Squadra: %(size)s su un massimo di %(max)s sottoagenti",
+        "How many work on it at once, the run's own agent among them, on the same model.": "Quanti ci lavorano alla volta, compreso l’agente dell’esecuzione, sullo stesso modello.",
         "Menu": "Menu",
         "Minimise": "Riduci a icona",
         "Maximise": "Ingrandisci",
@@ -1522,8 +1762,6 @@ OWN: dict[str, dict[str, str]] = {
         "A busy model is waited for, then the next free one is used.": "Si attende un modello occupato, poi si usa il successivo libero.",
         "A result, but not a run with nothing to report.": "Un risultato, ma non un’esecuzione senza nulla da segnalare.",
         "A task without a model of its own runs on the first free one of these, in this order. None ticked: all of them.": "Un’attività senza un modello proprio gira sul primo libero di questi, in quest’ordine. Nessuno spuntato: tutti.",
-        "A team of up to %(n)s": "Un team fino a %(n)s",
-        "A team works on it together, its own agent included.": "Un team ci lavora insieme, incluso il suo agente.",
         "Choose a model to choose its effort.": "Scegli un modello per sceglierne l’impegno.",
         "Due %(when)s": "Prevista %(when)s",
         "Each run sees the ones before it.": "Ogni esecuzione vede quelle precedenti.",
@@ -1532,7 +1770,6 @@ OWN: dict[str, dict[str, str]] = {
         "Edit %(title)s": "Modifica %(title)s",
         "Every result, even with nothing to report.": "Ogni risultato, anche senza nulla da segnalare.",
         "Models for scheduled tasks": "Modelli per le attività pianificate",
-        "No team": "Nessun team",
         "Only in its history.": "Solo nella sua cronologia.",
         "Open the chat": "Apri la chat",
         "Result": "Risultato",
@@ -1553,6 +1790,34 @@ OWN: dict[str, dict[str, str]] = {
         "No day ticked: every day.": "Nessun giorno spuntato: ogni giorno.",
     },
     "pl": {
+        "Sign in again: the server no longer accepts this app's sign-in.": "Zaloguj się ponownie: serwer nie przyjmuje już logowania tej aplikacji.",
+        "The terminal needs a newer LLeMbas server. Ask its administrator to update it.": "Terminal wymaga nowszego serwera LLeMbas. Poproś jego administratora o aktualizację.",
+        "This model serves one request at a time: subagents need another model, on another server, named in the chat.": "Ten model obsługuje jedno żądanie naraz: podagenci potrzebują innego modelu, na innym serwerze, wskazanego w czacie.",
+        "The numbers count this chat's own agent: 1 is the agent alone, with no subagents; Max 4 is the agent and up to 3 subagents. A subagent runs on the chat's model, or on another only when you name it in the chat.": "Liczby obejmują własnego agenta czatu: 1 to sam agent, bez podagentów; Maks 4 to agent i do 3 podagentów. Podagent działa na modelu czatu, na innym tylko wtedy, gdy wskażesz go w czacie.",
+        "Min": "Min",
+        "Max": "Maks",
+        "Tasks: %(n)s": "Zadania: %(n)s",
+        "%(n)s running": "Działa: %(n)s",
+        "%(tokens)s until auto-compact": "%(tokens)s do automatycznego streszczenia",
+        "Automatic compaction is off.": "Automatyczne streszczanie jest wyłączone.",
+        "Compacts before the next reply.": "Zostanie streszczona przed następną odpowiedzią.",
+        "Compacts by itself once the window is known.": "Zostanie streszczona sama, gdy okno będzie znane.",
+        "Compact": "Streść",
+        "Compacting…": "Streszczanie…",
+        "Summarise the earlier turns now and send the summary in their place.": "Streścić teraz wcześniejsze tury i wysyłać streszczenie zamiast nich.",
+        "This server cannot compact a chat from the app yet; the web interface can.": "Ten serwer nie umie jeszcze streścić czatu z aplikacji; interfejs WWW umie.",
+        "Context and background tasks": "Kontekst i zadania w tle",
+        "Environment": "Środowisko",
+        "Finished %(n)s": "Zakończone: %(n)s",
+        "Background command": "Polecenie w tle",
+        "Subagent": "Podagent",
+        "Team member": "Członek zespołu",
+        "Subagents": "Podagenci",
+        "None": "Brak",
+        "Up to %(n)s at once": "Do %(n)s naraz",
+        "Nothing runs in the background. Subagents, a team's members and background commands show here while they work.": "W tle nic nie działa. Podagenci, członkowie zespołu i polecenia w tle pojawiają się tu, gdy pracują.",
+        "Team: %(size)s of at most %(max)s subagents": "Zespół: %(size)s z najwyżej %(max)s podagentów",
+        "How many work on it at once, the run's own agent among them, on the same model.": "Ilu pracuje nad nim naraz, łącznie z własnym agentem uruchomienia, na tym samym modelu.",
         "Menu": "Menu",
         "Minimise": "Minimalizuj",
         "Maximise": "Maksymalizuj",
@@ -1720,8 +1985,6 @@ OWN: dict[str, dict[str, str]] = {
         "A busy model is waited for, then the next free one is used.": "Na zajęty model się czeka, potem używany jest następny wolny.",
         "A result, but not a run with nothing to report.": "Wynik, ale nie uruchomienie bez niczego do zgłoszenia.",
         "A task without a model of its own runs on the first free one of these, in this order. None ticked: all of them.": "Zadanie bez własnego modelu działa na pierwszym wolnym z tych, w tej kolejności. Żaden niezaznaczony: wszystkie.",
-        "A team of up to %(n)s": "Zespół do %(n)s",
-        "A team works on it together, its own agent included.": "Pracuje nad nim zespół, wraz z jego własnym agentem.",
         "Choose a model to choose its effort.": "Wybierz model, aby wybrać jego wysiłek.",
         "Due %(when)s": "Zaplanowane na %(when)s",
         "Each run sees the ones before it.": "Każde uruchomienie widzi poprzednie.",
@@ -1730,7 +1993,6 @@ OWN: dict[str, dict[str, str]] = {
         "Edit %(title)s": "Edytuj %(title)s",
         "Every result, even with nothing to report.": "Każdy wynik, nawet bez niczego do zgłoszenia.",
         "Models for scheduled tasks": "Modele dla zaplanowanych zadań",
-        "No team": "Bez zespołu",
         "Only in its history.": "Tylko w jego historii.",
         "Open the chat": "Otwórz czat",
         "Result": "Wynik",
@@ -1751,6 +2013,34 @@ OWN: dict[str, dict[str, str]] = {
         "No day ticked: every day.": "Żaden dzień niezaznaczony: codziennie.",
     },
     "sk": {
+        "Sign in again: the server no longer accepts this app's sign-in.": "Prihláste sa znova: server už neprijíma prihlásenie tejto aplikácie.",
+        "The terminal needs a newer LLeMbas server. Ask its administrator to update it.": "Terminál potrebuje novší server LLeMbas. Požiadajte jeho správcu o aktualizáciu.",
+        "This model serves one request at a time: subagents need another model, on another server, named in the chat.": "Tento model vybavuje naraz jednu požiadavku: podagenti potrebujú iný model na inom serveri, pomenovaný v konverzácii.",
+        "The numbers count this chat's own agent: 1 is the agent alone, with no subagents; Max 4 is the agent and up to 3 subagents. A subagent runs on the chat's model, or on another only when you name it in the chat.": "Čísla počítajú aj vlastného agenta konverzácie: 1 je agent sám, bez podagentov; Max 4 je agent a až 3 podagenti. Podagent beží na modeli konverzácie, na inom len vtedy, keď ho v konverzácii pomenujete.",
+        "Min": "Min",
+        "Max": "Max",
+        "Tasks: %(n)s": "Úlohy: %(n)s",
+        "%(n)s running": "Beží: %(n)s",
+        "%(tokens)s until auto-compact": "%(tokens)s do automatického zhrnutia",
+        "Automatic compaction is off.": "Automatické zhrnutie je vypnuté.",
+        "Compacts before the next reply.": "Zhrnie sa pred ďalšou odpoveďou.",
+        "Compacts by itself once the window is known.": "Zhrnie sa samo, keď bude okno známe.",
+        "Compact": "Zhrnúť",
+        "Compacting…": "Zhŕňa sa…",
+        "Summarise the earlier turns now and send the summary in their place.": "Zhrnúť skoršie kolá hneď a posielať namiesto nich zhrnutie.",
+        "This server cannot compact a chat from the app yet; the web interface can.": "Tento server zatiaľ nevie zhrnúť konverzáciu z aplikácie; webové rozhranie to vie.",
+        "Context and background tasks": "Kontext a úlohy na pozadí",
+        "Environment": "Prostredie",
+        "Finished %(n)s": "Dokončené: %(n)s",
+        "Background command": "Príkaz na pozadí",
+        "Subagent": "Podagent",
+        "Team member": "Člen tímu",
+        "Subagents": "Podagenti",
+        "None": "Žiadni",
+        "Up to %(n)s at once": "Najviac %(n)s naraz",
+        "Nothing runs in the background. Subagents, a team's members and background commands show here while they work.": "Na pozadí nič nebeží. Podagenti, členovia tímu a príkazy na pozadí sa tu zobrazia, kým pracujú.",
+        "Team: %(size)s of at most %(max)s subagents": "Tím: %(size)s z najviac %(max)s podagentov",
+        "How many work on it at once, the run's own agent among them, on the same model.": "Koľko na nej pracuje naraz, vrátane vlastného agenta behu, na rovnakom modeli.",
         "Menu": "Ponuka",
         "Minimise": "Minimalizovať",
         "Maximise": "Maximalizovať",
@@ -1918,8 +2208,6 @@ OWN: dict[str, dict[str, str]] = {
         "A busy model is waited for, then the next free one is used.": "Na zaneprázdnený model sa čaká, potom sa použije ďalší voľný.",
         "A result, but not a run with nothing to report.": "Výsledok, ale nie spustenie, ktoré nemá čo hlásiť.",
         "A task without a model of its own runs on the first free one of these, in this order. None ticked: all of them.": "Úloha bez vlastného modelu beží na prvom voľnom z týchto, v tomto poradí. Žiadny nezaškrtnutý: všetky.",
-        "A team of up to %(n)s": "Tím: najviac %(n)s",
-        "A team works on it together, its own agent included.": "Pracuje na nej tím spoločne, vrátane jej vlastného agenta.",
         "Choose a model to choose its effort.": "Vyberte model, aby ste mohli zvoliť jeho náročnosť.",
         "Due %(when)s": "Naplánované na %(when)s",
         "Each run sees the ones before it.": "Každé spustenie vidí tie predchádzajúce.",
@@ -1928,7 +2216,6 @@ OWN: dict[str, dict[str, str]] = {
         "Edit %(title)s": "Upraviť %(title)s",
         "Every result, even with nothing to report.": "Každý výsledok, aj keď nie je čo hlásiť.",
         "Models for scheduled tasks": "Modely pre naplánované úlohy",
-        "No team": "Bez tímu",
         "Only in its history.": "Len v jej histórii.",
         "Open the chat": "Otvoriť konverzáciu",
         "Result": "Výsledok",
@@ -1997,7 +2284,7 @@ def main() -> None:
                 print(f"gen-i18n: {lang}: OWN has {key!r}, which no page uses", file=sys.stderr)
                 problems += 1
         catalogue[lang] = table
-        print(f"gen-i18n: {lang}: {len(table)} of {len(KEYS)} ({from_server} from the server)")
+        print(f"gen-i18n: {lang}: {len(table)} sentences, {from_server} of the page's {len(KEYS)} from the server")
 
     body = json.dumps(catalogue, ensure_ascii=False, indent=2, sort_keys=True)
     out = (

src-tauri/Cargo.lock

@@ -279,6 +279,15 @@ dependencies = [
  "generic-array",
 ]
 
+[[package]]
+name = "block-buffer"
+version = "0.12.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "d2f6c7dbe95a6ed67ad9f18e57daf93a2f034c524b99fd2b76d18fdfeb6660aa"
+dependencies = [
+ "hybrid-array",
+]
+
 [[package]]
 name = "block2"
 version = "0.6.2"
@@ -511,6 +520,12 @@ dependencies = [
  "crossbeam-utils",
 ]
 
+[[package]]
+name = "const-oid"
+version = "0.10.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "a6ef517f0926dd24a1582492c791b6a4818a4d94e789a334894aa15b0d12f55c"
+
 [[package]]
 name = "cookie"
 version = "0.18.2"
@@ -623,6 +638,15 @@ dependencies = [
  "typenum",
 ]
 
+[[package]]
+name = "crypto-common"
+version = "0.2.2"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "ce6e4c961d6cd6c9a86db418387425e8bdeaf05b3c8bc1411e6dca4c252f1453"
+dependencies = [
+ "hybrid-array",
+]
+
 [[package]]
 name = "cssparser"
 version = "0.37.0"
@@ -686,6 +710,12 @@ dependencies = [
  "syn 3.0.8",
 ]
 
+[[package]]
+name = "data-encoding"
+version = "2.11.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "4583a4551df46e2792f82ceeac45e850d2e2d5debba0b91f102385cda5b11f06"
+
 [[package]]
 name = "dbus"
 version = "0.9.12"
@@ -775,8 +805,19 @@ version = "0.10.7"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "9ed9a281f7bc9b7576e61468ba615a66a5c8cfdff42420a70aa82701a3b1e292"
 dependencies = [
- "block-buffer",
+ "block-buffer 0.10.4",
- "crypto-common",
+ "crypto-common 0.1.7",
+]
+
+[[package]]
+name = "digest"
+version = "0.11.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f1dd6dbb5841937940781866fa1281a1ff7bd3bf827091440879f9994983d5c2"
+dependencies = [
+ "block-buffer 0.12.1",
+ "const-oid",
+ "crypto-common 0.2.2",
 ]
 
 [[package]]
@@ -1552,6 +1593,15 @@ version = "1.10.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "6dbf3de79e51f3d586ab4cb9d5c3e2c14aa28ed23d180cf89b4df0454a69cc87"
 
+[[package]]
+name = "hybrid-array"
+version = "0.4.15"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "27f864f10dfb56725ce5ce5472bc52252c8f93a4ab86327122cebf62c5f59a17"
+dependencies = [
+ "typenum",
+]
+
 [[package]]
 name = "hyper"
 version = "1.12.0"
@@ -2085,7 +2135,7 @@ checksum = "47d9d19d1d6efa0109d2f65ff4c85cddd50bd572e5a00127ab10987290bcefae"
 
 [[package]]
 name = "llembas-desktop"
-version = "1.0.0-beta.4"
+version = "1.0.0-beta.6"
 dependencies = [
  "libc",
  "serde",
@@ -2097,6 +2147,7 @@ dependencies = [
  "tauri-plugin-opener",
  "tauri-plugin-single-instance",
  "tauri-plugin-updater",
+ "tungstenite",
  "ureq",
 ]
 
@@ -3488,6 +3539,17 @@ dependencies = [
  "stable_deref_trait",
 ]
 
+[[package]]
+name = "sha1"
+version = "0.11.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "aacc4cc499359472b4abe1bf11d0b12e688af9a805fa5e3016f9a386dc2d0214"
+dependencies = [
+ "cfg-if",
+ "cpufeatures 0.3.1",
+ "digest 0.11.3",
+]
+
 [[package]]
 name = "sha2"
 version = "0.10.9"
@@ -3496,7 +3558,7 @@ checksum = "a7507d819769d01a365ab707794a4084392c824f54a7a6a7862f8c3d0892b283"
 dependencies = [
  "cfg-if",
  "cpufeatures 0.2.17",
- "digest",
+ "digest 0.10.7",
 ]
 
 [[package]]
@@ -4539,6 +4601,25 @@ version = "0.2.5"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "e421abadd41a4225275504ea4d6566923418b7f05506fbc9c0fe86ba7396114b"
 
+[[package]]
+name = "tungstenite"
+version = "0.30.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e48ac77174b19c110a50ab2128b24215ac9cb40e0e12e093fb602d175c569d22"
+dependencies = [
+ "bytes",
+ "data-encoding",
+ "http",
+ "httparse",
+ "log",
+ "rand",
+ "rustls",
+ "rustls-native-certs 0.8.5",
+ "rustls-pki-types",
+ "sha1",
+ "thiserror 2.0.21",
+]
+
 [[package]]
 name = "typeid"
 version = "1.0.3"

src-tauri/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "llembas-desktop"
-version = "1.0.0-beta.4"
+version = "1.0.0-beta.6"
 description = "LLeMbas Desktop: the agent on this computer and your LLeMbas server's chats, in an app of its own"
 authors = ["Jaroslav Beneš"]
 license = "MIT"
@@ -26,6 +26,8 @@ tauri-plugin-updater = "2"
 serde = { version = "1", features = ["derive"] }
 serde_json = "1"
 ureq = { version = "2", default-features = false, features = ["tls", "native-certs", "json"] }
+# The terminal panel's socket (`terminal.rs`): a blocking WebSocket on the TLS ureq already brings.
+tungstenite = { version = "0.30", default-features = false, features = ["handshake", "rustls-tls-native-roots"] }
 
 [target.'cfg(unix)'.dependencies]
 libc = "0.2"

src-tauri/src/lib.rs

@@ -22,6 +22,7 @@ mod remote;
 mod sealed;
 mod server;
 mod settings;
+mod terminal;
 mod watch;
 
 use std::collections::HashMap;
@@ -115,6 +116,8 @@ struct App {
     signing_in: Mutex<Option<Arc<AtomicBool>>>,
     streams: Mutex<HashMap<u64, Arc<AtomicBool>>>,
     next_stream: AtomicU64,
+    /// The terminal panel's sockets (`terminal.rs`), by the page's id for each.
+    terminals: terminal::Sessions,
     sealed: sealed::Pages,
     watch: watch::Watch,
     /// The files last dropped on the window: what `attach_dropped` may send, and nothing else.
@@ -271,6 +274,7 @@ async fn set_server(handle: AppHandle, address: String) -> Result<Status, String
     if before.as_deref() != Some(d.base.as_str()) {
         app.agent.stop();
         app.watch.reset();
+        app.terminals.close_all();
     }
     *app.account.lock().unwrap() = server::load_account(&app.config_dir, &d.base);
     *app.server.lock().unwrap() = Some(d);
@@ -380,6 +384,7 @@ fn sign_out(handle: AppHandle) {
     let account = app.account.lock().unwrap().take();
     server::forget_account(&app.config_dir);
     app.watch.reset();
+    app.terminals.close_all();
     if let Some(a) = account {
         std::thread::spawn(move || server::sign_out(&a.server, &a.token));
     }
@@ -391,6 +396,7 @@ fn token_refused(handle: &AppHandle) {
     let app = handle.state::<App>();
     if app.account.lock().unwrap().take().is_some() {
         server::forget_account(&app.config_dir);
+        app.terminals.close_all();
         announce(handle);
     }
 }
@@ -511,6 +517,71 @@ fn stream_close(app: tauri::State<'_, App>, id: u64) {
     }
 }
 
+// --- The terminal panel ------------------------------------------------------------------------
+
+/// Open a shell on a linked device, as the terminal panel asks: an agent chat's (`chat`), or the
+/// one before the chat exists (`device` and `directory`). The socket is opened here, with the
+/// app's token in its handshake; the page gets its frames as `terminal` events with this id --
+/// `output` (base64), `control` (the server's JSON), and a last `end`.
+#[tauri::command]
+fn terminal_open(
+    handle: AppHandle,
+    chat: Option<String>,
+    device: Option<String>,
+    directory: Option<String>,
+    cols: u16,
+    rows: u16,
+) -> Result<u64, ApiError> {
+    let app = handle.state::<App>();
+    let Some((app_url, token)) = app.door() else {
+        return Err(ApiError::new(401, "Not signed in."));
+    };
+    let target = match (chat, device) {
+        (Some(chat), None) => terminal::Target::Chat(chat),
+        (None, Some(device)) => terminal::Target::Device {
+            device,
+            directory: directory.unwrap_or_default(),
+        },
+        _ => return Err(ApiError::new(400, "A chat or a device, not both.")),
+    };
+    let url = terminal::url(&app_url, &target, cols, rows)?;
+    let h = handle.clone();
+    let on_end = handle.clone();
+    Ok(app.terminals.open(
+        url,
+        token,
+        move |down| {
+            let _ = h.emit("terminal", down);
+        },
+        move |id, end| {
+            on_end.state::<App>().terminals.forget(id);
+            if end.unauthorized() {
+                token_refused(&on_end);
+            }
+        },
+    ))
+}
+
+/// Keys typed (or pasted) into a terminal, as text: sent to the shell as its bytes.
+#[tauri::command]
+fn terminal_input(app: tauri::State<'_, App>, id: u64, data: String) -> bool {
+    app.terminals
+        .send(id, terminal::Up::Keys(data.into_bytes()))
+}
+
+/// The terminal's size changed.
+#[tauri::command]
+fn terminal_resize(app: tauri::State<'_, App>, id: u64, cols: u16, rows: u16) -> bool {
+    app.terminals.send(id, terminal::Up::Resize(cols, rows))
+}
+
+/// The page let go of a terminal: its socket is closed (the shell on the device lives on, as
+/// the server keeps it).
+#[tauri::command]
+fn terminal_close(app: tauri::State<'_, App>, id: u64) {
+    app.terminals.close(id);
+}
+
 // --- Attachments ----------------------------------------------------------------------------------
 
 /// What became of one file sent as an attachment: the server's attachment, or why not. `reason`
@@ -1560,6 +1631,10 @@ pub fn run() {
             terminal_word,
             stream_open,
             stream_close,
+            terminal_open,
+            terminal_input,
+            terminal_resize,
+            terminal_close,
             local_list,
             local_read,
             sealed_open,
@@ -1621,6 +1696,7 @@ pub fn run() {
                 signing_in: Mutex::new(None),
                 streams: Mutex::new(HashMap::new()),
                 next_stream: AtomicU64::new(0),
+                terminals: terminal::Sessions::default(),
                 sealed: sealed::Pages::default(),
                 watch: watch::Watch::default(),
                 dropped: Mutex::new(Vec::new()),

src-tauri/src/terminal.rs

@@ -0,0 +1,444 @@
+//! The terminal panel's socket: a shell on a linked device, through the server's app door.
+//!
+//! The server has two WebSockets for an app (`/api/v2`, behind the app's token):
+//!
+//! - `/chats/{chat}/terminal/ws` -- an agent chat's shell, the one the web page's panel opens
+//!   too, so the two share it;
+//! - `/devices/{device}/terminal/ws?directory=` -- before the chat exists, on the device and in
+//!   the folder a new agent chat has chosen. The chat's first message takes that shell over.
+//!
+//! Both take `cols` and `rows`. Down: the shell's output as binary frames, and text frames
+//! `{"t": "ready" | "closed" | "behind" | "error", ...}`. Up: keystrokes as binary,
+//! `{"t": "resize", "cols", "rows"}` as text. A refused token closes with 4401.
+//!
+//! **The token stays here.** The socket is opened in Rust with `Authorization: Bearer` in the
+//! handshake, and the page only ever gets the frames (as `terminal` events) and sends keys,
+//! sizes and "close" through commands -- the same rule as every other request the page makes.
+//! A page cannot open the socket itself: its content security policy allows it no connection
+//! but the bridge.
+//!
+//! One thread per terminal owns its socket: it reads with a short timeout and, between reads,
+//! sends what the page queued on its channel. Blocking sockets, no async runtime of its own.
+
+use std::collections::HashMap;
+use std::io::ErrorKind;
+use std::net::{TcpStream, ToSocketAddrs};
+use std::sync::atomic::{AtomicU64, Ordering};
+use std::sync::mpsc::{self, Receiver, Sender, TryRecvError};
+use std::sync::Mutex;
+use std::time::{Duration, Instant};
+
+use serde::Serialize;
+use serde_json::{json, Value};
+use tungstenite::http::Uri;
+use tungstenite::stream::MaybeTlsStream;
+use tungstenite::{ClientRequestBuilder, Error as WsError, HandshakeError, Message, WebSocket};
+
+use crate::server::{base64, ApiError};
+
+/// The server's close code for a token it refused (`api/v2/terminal.py`).
+pub const CLOSE_UNAUTHORIZED: u16 = 4401;
+/// The most one keystroke message may be: the server's own limit (a paste, at most).
+pub const MAX_INPUT: usize = 256 * 1024;
+/// The longest directory the server takes (`api/terminal.py` `MAX_DIR_CHARS`).
+const MAX_DIR: usize = 4096;
+/// How long a read waits before the thread looks at what the page queued.
+const POLL: Duration = Duration::from_millis(40);
+/// How long a close waits for the server's answer before the socket is simply dropped.
+const CLOSE_WAIT: Duration = Duration::from_secs(2);
+
+type Socket = WebSocket<MaybeTlsStream<TcpStream>>;
+
+/// Where the shell is: an agent chat's, or a device's folder before the chat exists.
+#[derive(Clone, Debug, PartialEq)]
+pub enum Target {
+    Chat(String),
+    Device { device: String, directory: String },
+}
+
+/// An id as the server makes them (a chat's, a device token's): letters, digits, `-` and `_`.
+fn id_ok(id: &str) -> bool {
+    !id.is_empty()
+        && id.len() <= 128
+        && id
+            .chars()
+            .all(|c| c.is_ascii_alphanumeric() || c == '-' || c == '_')
+}
+
+/// A query value, percent-encoded: every byte but the unreserved ones (RFC 3986 §2.3).
+pub fn encode(text: &str) -> String {
+    let mut out = String::with_capacity(text.len());
+    for b in text.bytes() {
+        if b.is_ascii_alphanumeric() || matches!(b, b'-' | b'_' | b'.' | b'~') {
+            out.push(b as char);
+        } else {
+            out.push_str(&format!("%{b:02X}"));
+        }
+    }
+    out
+}
+
+/// The socket's address on the app door (`https://host/api/v2` -> `wss://host/api/v2/...`).
+pub fn url(app: &str, target: &Target, cols: u16, rows: u16) -> Result<String, ApiError> {
+    let door = if let Some(rest) = app.strip_prefix("https://") {
+        format!("wss://{rest}")
+    } else if let Some(rest) = app.strip_prefix("http://") {
+        format!("ws://{rest}")
+    } else {
+        return Err(ApiError::new(
+            400,
+            "The server's address is not http or https.",
+        ));
+    };
+    let cols = cols.clamp(2, 1000);
+    let rows = rows.clamp(1, 500);
+    match target {
+        Target::Chat(chat) => {
+            if !id_ok(chat) {
+                return Err(ApiError::new(400, "Not a chat."));
+            }
+            Ok(format!(
+                "{door}/chats/{chat}/terminal/ws?cols={cols}&rows={rows}"
+            ))
+        }
+        Target::Device { device, directory } => {
+            if !id_ok(device) {
+                return Err(ApiError::new(400, "Not a device."));
+            }
+            let directory = directory.trim();
+            if directory.is_empty() || directory.len() > MAX_DIR || directory.contains('\0') {
+                return Err(ApiError::new(400, "Not a folder."));
+            }
+            Ok(format!(
+                "{door}/devices/{device}/terminal/ws?directory={}&cols={cols}&rows={rows}",
+                encode(directory)
+            ))
+        }
+    }
+}
+
+/// What the page asked the socket's thread to do.
+pub enum Up {
+    Keys(Vec<u8>),
+    Resize(u16, u16),
+    Close,
+}
+
+/// One frame for the page, as the `terminal` event: `output` (the bytes, base64), `control`
+/// (the server's JSON), or `end` (how it ended: `{code, reason, status, error}`).
+#[derive(Clone, Serialize)]
+pub struct Down {
+    pub id: u64,
+    pub kind: &'static str,
+    pub data: Value,
+}
+
+/// How a terminal's socket ended. `code` is the close frame's (4401: the token was refused),
+/// `status` an HTTP status the handshake was answered with, `error` why it could not connect or
+/// broke off.
+#[derive(Clone, Debug, Default, PartialEq, Serialize)]
+pub struct End {
+    pub code: Option<u16>,
+    pub reason: String,
+    pub status: Option<u16>,
+    pub error: Option<String>,
+}
+
+impl End {
+    fn failed(error: impl Into<String>) -> Self {
+        End {
+            error: Some(error.into()),
+            ..End::default()
+        }
+    }
+
+    /// The server refused the app's token: the page signs in again.
+    pub fn unauthorized(&self) -> bool {
+        self.code == Some(CLOSE_UNAUTHORIZED) || self.status == Some(401)
+    }
+}
+
+/// The open terminals: what each one's thread listens to.
+#[derive(Default)]
+pub struct Sessions {
+    next: AtomicU64,
+    open: Mutex<HashMap<u64, Sender<Up>>>,
+}
+
+impl Sessions {
+    /// Open a terminal's socket on its own thread. `emit` gets every frame, the `end` last;
+    /// `ended` is told how it ended just before. The id is the page's handle on it.
+    pub fn open(
+        &self,
+        url: String,
+        token: String,
+        emit: impl Fn(Down) + Send + 'static,
+        ended: impl FnOnce(u64, &End) + Send + 'static,
+    ) -> u64 {
+        let id = self.next.fetch_add(1, Ordering::SeqCst) + 1;
+        let (tx, rx) = mpsc::channel();
+        self.open.lock().unwrap().insert(id, tx);
+        std::thread::spawn(move || {
+            let send = |kind: &'static str, data: Value| emit(Down { id, kind, data });
+            let end = run(&url, &token, &rx, &send);
+            ended(id, &end);
+            send("end", serde_json::to_value(&end).unwrap_or(Value::Null));
+        });
+        id
+    }
+
+    /// Hand the page's keys, size or close to the terminal's thread. False: no such terminal.
+    pub fn send(&self, id: u64, up: Up) -> bool {
+        self.open
+            .lock()
+            .unwrap()
+            .get(&id)
+            .is_some_and(|tx| tx.send(up).is_ok())
+    }
+
+    /// Close one terminal (the page's panel let go of it).
+    pub fn close(&self, id: u64) {
+        if let Some(tx) = self.open.lock().unwrap().remove(&id) {
+            let _ = tx.send(Up::Close);
+        }
+    }
+
+    /// A terminal's thread has ended: forget it.
+    pub fn forget(&self, id: u64) {
+        self.open.lock().unwrap().remove(&id);
+    }
+
+    /// Every terminal closed (signed out, or another server chosen).
+    pub fn close_all(&self) {
+        for (_, tx) in self.open.lock().unwrap().drain() {
+            let _ = tx.send(Up::Close);
+        }
+    }
+}
+
+/// The TCP socket under the WebSocket, to set its read timeout.
+fn tcp(ws: &Socket) -> Option<&TcpStream> {
+    match ws.get_ref() {
+        MaybeTlsStream::Plain(s) => Some(s),
+        MaybeTlsStream::Rustls(s) => Some(s.get_ref()),
+        _ => None,
+    }
+}
+
+fn connect(url: &str, token: &str) -> Result<Socket, End> {
+    let uri: Uri = url
+        .parse()
+        .map_err(|e| End::failed(format!("Not an address: {e}")))?;
+    let host = uri
+        .host()
+        .ok_or_else(|| End::failed("The address has no host."))?
+        .trim_start_matches('[')
+        .trim_end_matches(']')
+        .to_string();
+    let port = uri
+        .port_u16()
+        .unwrap_or(if uri.scheme_str() == Some("wss") {
+            443
+        } else {
+            80
+        });
+    let addrs = (host.as_str(), port)
+        .to_socket_addrs()
+        .map_err(|e| End::failed(format!("Could not reach the server: {e}")))?;
+    let mut last = String::from("no address");
+    let mut stream = None;
+    for addr in addrs {
+        match TcpStream::connect_timeout(&addr, Duration::from_secs(10)) {
+            Ok(s) => {
+                stream = Some(s);
+                break;
+            }
+            Err(e) => last = e.to_string(),
+        }
+    }
+    let stream =
+        stream.ok_or_else(|| End::failed(format!("Could not reach the server: {last}")))?;
+    // The handshake may take its time; reads only poll once it is done.
+    let _ = stream.set_read_timeout(Some(Duration::from_secs(30)));
+    let _ = stream.set_nodelay(true);
+    let request = ClientRequestBuilder::new(uri)
+        .with_header("Authorization", format!("Bearer {token}"))
+        .with_header(
+            "User-Agent",
+            format!("llembas-desktop/{}", env!("CARGO_PKG_VERSION")),
+        );
+    match tungstenite::client_tls(request, stream) {
+        Ok((ws, _)) => Ok(ws),
+        Err(HandshakeError::Failure(WsError::Http(res))) => Err(End {
+            status: Some(res.status().as_u16()),
+            ..End::default()
+        }),
+        Err(HandshakeError::Failure(e)) => Err(End::failed(format!("Could not connect: {e}"))),
+        Err(HandshakeError::Interrupted(_)) => Err(End::failed("The server did not answer.")),
+    }
+}
+
+/// A read that only timed out: nothing came in this poll.
+fn idle(e: &WsError) -> bool {
+    matches!(e, WsError::Io(io) if matches!(io.kind(), ErrorKind::WouldBlock | ErrorKind::TimedOut))
+}
+
+fn run(url: &str, token: &str, rx: &Receiver<Up>, send: &dyn Fn(&'static str, Value)) -> End {
+    let mut ws = match connect(url, token) {
+        Ok(ws) => ws,
+        Err(end) => return end,
+    };
+    match tcp(&ws) {
+        Some(s) => {
+            let _ = s.set_read_timeout(Some(POLL));
+        }
+        None => return End::failed("An unexpected kind of connection."),
+    }
+    let mut closed_by_server: Option<(u16, String)> = None;
+    let mut closing: Option<Instant> = None;
+    loop {
+        // What the page queued, until the server has said goodbye.
+        if closing.is_none() && closed_by_server.is_none() {
+            loop {
+                let sent = match rx.try_recv() {
+                    Ok(Up::Keys(bytes)) if bytes.len() <= MAX_INPUT => {
+                        ws.send(Message::binary(bytes))
+                    }
+                    Ok(Up::Keys(_)) => Ok(()),
+                    Ok(Up::Resize(cols, rows)) => ws.send(Message::text(
+                        json!({ "t": "resize", "cols": cols, "rows": rows }).to_string(),
+                    )),
+                    Ok(Up::Close) | Err(TryRecvError::Disconnected) => {
+                        closing = Some(Instant::now());
+                        let _ = ws.close(None);
+                        break;
+                    }
+                    Err(TryRecvError::Empty) => break,
+                };
+                if let Err(e) = sent {
+                    if !idle(&e) {
+                        return End::failed(format!("The connection broke off: {e}"));
+                    }
+                }
+            }
+        }
+        match ws.read() {
+            Ok(Message::Binary(bytes)) => send("output", Value::String(base64(&bytes))),
+            Ok(Message::Text(text)) => {
+                if let Ok(value) = serde_json::from_str::<Value>(text.as_str()) {
+                    send("control", value);
+                }
+            }
+            Ok(Message::Close(frame)) => {
+                let (code, reason) = frame
+                    .map(|f| (u16::from(f.code), f.reason.as_str().to_string()))
+                    .unwrap_or((1005, String::new()));
+                closed_by_server = Some((code, reason));
+            }
+            Ok(_) => {}
+            Err(e) if idle(&e) => {
+                if closing.is_some_and(|at| at.elapsed() > CLOSE_WAIT) {
+                    break;
+                }
+            }
+            Err(WsError::ConnectionClosed) | Err(WsError::AlreadyClosed) => break,
+            Err(e) => {
+                if closing.is_some() || closed_by_server.is_some() {
+                    break;
+                }
+                return End::failed(format!("The connection broke off: {e}"));
+            }
+        }
+    }
+    match closed_by_server {
+        Some((code, reason)) => End {
+            code: Some(code),
+            reason,
+            ..End::default()
+        },
+        None => End::default(),
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn a_chats_terminal_is_on_the_app_door_with_its_size() {
+        let t = Target::Chat("chat_1-a".into());
+        assert_eq!(
+            url("https://ai.example/api/v2", &t, 80, 24).unwrap(),
+            "wss://ai.example/api/v2/chats/chat_1-a/terminal/ws?cols=80&rows=24"
+        );
+        assert_eq!(
+            url("http://ai.example:8080/sub/api/v2", &t, 0, 9999).unwrap(),
+            "ws://ai.example:8080/sub/api/v2/chats/chat_1-a/terminal/ws?cols=2&rows=500"
+        );
+    }
+
+    #[test]
+    fn a_terminal_before_the_chat_carries_its_folder_encoded() {
+        let t = Target::Device {
+            device: "token1".into(),
+            directory: " /home/dev/my project&x=1 ".into(),
+        };
+        assert_eq!(
+            url("https://ai.example/api/v2", &t, 100, 30).unwrap(),
+            "wss://ai.example/api/v2/devices/token1/terminal/ws?directory=%2Fhome%2Fdev%2Fmy%20project%26x%3D1&cols=100&rows=30"
+        );
+        assert_eq!(encode("C:\\Users\\Ádám"), "C%3A%5CUsers%5C%C3%81d%C3%A1m");
+    }
+
+    #[test]
+    fn nothing_but_an_id_and_a_folder_reaches_the_address() {
+        let door = "https://ai.example/api/v2";
+        for bad in ["", "../me", "a/b", "a?b", "a b", "x#y"] {
+            assert!(
+                url(door, &Target::Chat(bad.into()), 80, 24).is_err(),
+                "{bad}"
+            );
+            let dev = Target::Device {
+                device: bad.into(),
+                directory: "/tmp".into(),
+            };
+            assert!(url(door, &dev, 80, 24).is_err(), "{bad}");
+        }
+        for dir in ["", "   ", "a\0b"] {
+            let dev = Target::Device {
+                device: "token1".into(),
+                directory: dir.into(),
+            };
+            assert!(url(door, &dev, 80, 24).is_err(), "{dir:?}");
+        }
+        assert!(url("ftp://ai.example", &Target::Chat("c".into()), 80, 24).is_err());
+    }
+
+    #[test]
+    fn a_refused_token_is_told_apart_from_a_dropped_line() {
+        let refused = End {
+            code: Some(CLOSE_UNAUTHORIZED),
+            ..End::default()
+        };
+        assert!(refused.unauthorized());
+        assert!(End {
+            status: Some(401),
+            ..End::default()
+        }
+        .unauthorized());
+        assert!(!End::failed("reset").unauthorized());
+        assert!(!End {
+            code: Some(1000),
+            ..End::default()
+        }
+        .unauthorized());
+    }
+
+    #[test]
+    fn keys_for_a_closed_terminal_go_nowhere() {
+        let s = Sessions::default();
+        assert!(!s.send(7, Up::Keys(b"ls\n".to_vec())));
+        s.close(7);
+        s.close_all();
+    }
+}

tests/ui/conformance/progress.json

@@ -0,0 +1,333 @@
+{
+  "about": "Progress (progress.md, spec 3.4.0): the text and the line every client draws for a progress object. `progress`: the object; `now`: the client's clock (Unix ms). Expect `text` — the phase's text from loop.json progress.texts, its fields filled (numbers written plainly: integers as they are, a rate with one decimal) — and `line` — `{{text}} · {{elapsed}}` with elapsed = now − since in whole seconds (5s, 1m 05s, 1h 02m), or the text alone for stopped and failed. retrying's `in` is counted down: in − floor((now − since) / 1000), never below 0. LLeMbas: services/progress.py; CLI: session/progress.ts.",
+  "cases": [
+    {
+      "about": "Queued, nothing more known.",
+      "progress": {
+        "phase": "queued",
+        "since": 1762790400000
+      },
+      "now": 1762790405000,
+      "expect": {
+        "text": "Queued",
+        "line": "Queued · 5s"
+      }
+    },
+    {
+      "about": "Queued behind two turns.",
+      "progress": {
+        "phase": "queued",
+        "since": 1762790400000,
+        "ahead": 2
+      },
+      "now": 1762790465000,
+      "expect": {
+        "text": "Queued: 2 ahead",
+        "line": "Queued: 2 ahead · 1m 05s"
+      }
+    },
+    {
+      "about": "A one-request server answering another request.",
+      "progress": {
+        "phase": "waiting_for_model",
+        "since": 1762790400000,
+        "model": "Qwen 27B",
+        "reason": "busy"
+      },
+      "now": 1762790412400,
+      "expect": {
+        "text": "Waiting for Qwen 27B: it serves one request at a time and is answering another",
+        "line": "Waiting for Qwen 27B: it serves one request at a time and is answering another · 12s"
+      }
+    },
+    {
+      "about": "Every slot taken.",
+      "progress": {
+        "phase": "waiting_for_model",
+        "since": 1762790400000,
+        "model": "coder",
+        "reason": "slots"
+      },
+      "now": 1762790401000,
+      "expect": {
+        "text": "Waiting for coder: all its slots are in use",
+        "line": "Waiting for coder: all its slots are in use · 1s"
+      }
+    },
+    {
+      "about": "Another model on the one-model server.",
+      "progress": {
+        "phase": "waiting_for_model",
+        "since": 1762790400000,
+        "model": "coder",
+        "reason": "group"
+      },
+      "now": 1762790400000,
+      "expect": {
+        "text": "Waiting for coder: its server holds another model now",
+        "line": "Waiting for coder: its server holds another model now · 0s"
+      }
+    },
+    {
+      "about": "Loading.",
+      "progress": {
+        "phase": "loading_model",
+        "since": 1762790400000,
+        "model": "Gemma 12B"
+      },
+      "now": 1762794125000,
+      "expect": {
+        "text": "Loading Gemma 12B",
+        "line": "Loading Gemma 12B · 1h 02m"
+      }
+    },
+    {
+      "about": "Reading the prompt: the percentage is floored.",
+      "progress": {
+        "phase": "reading_prompt",
+        "since": 1762790400000,
+        "done": 1999,
+        "total": 3000
+      },
+      "now": 1762790402000,
+      "expect": {
+        "text": "Reading the prompt: 1999 / 3000 tokens (66%)",
+        "line": "Reading the prompt: 1999 / 3000 tokens (66%) · 2s"
+      }
+    },
+    {
+      "about": "Reading the prompt with no counts from the server.",
+      "progress": {
+        "phase": "reading_prompt",
+        "since": 1762790400000
+      },
+      "now": 1762790402000,
+      "expect": {
+        "text": "Reading the prompt",
+        "line": "Reading the prompt · 2s"
+      }
+    },
+    {
+      "about": "Thinking.",
+      "progress": {
+        "phase": "thinking",
+        "since": 1762790400000
+      },
+      "now": 1762790459999,
+      "expect": {
+        "text": "Thinking",
+        "line": "Thinking · 59s"
+      }
+    },
+    {
+      "about": "Writing, the rate with one decimal.",
+      "progress": {
+        "phase": "writing",
+        "since": 1762790400000,
+        "rate": 12.46
+      },
+      "now": 1762790404000,
+      "expect": {
+        "text": "Writing: 12.5 tok/s",
+        "line": "Writing: 12.5 tok/s · 4s"
+      }
+    },
+    {
+      "about": "Writing before a rate is measured.",
+      "progress": {
+        "phase": "writing",
+        "since": 1762790400000
+      },
+      "now": 1762790400400,
+      "expect": {
+        "text": "Writing",
+        "line": "Writing · 0s"
+      }
+    },
+    {
+      "about": "A tool with what it does.",
+      "progress": {
+        "phase": "running_tool",
+        "since": 1762790400000,
+        "tool": "bash",
+        "what": "bun test"
+      },
+      "now": 1762790461000,
+      "expect": {
+        "text": "Running bash: bun test",
+        "line": "Running bash: bun test · 1m 01s"
+      }
+    },
+    {
+      "about": "A tool without a title.",
+      "progress": {
+        "phase": "running_tool",
+        "since": 1762790400000,
+        "tool": "recall"
+      },
+      "now": 1762790401000,
+      "expect": {
+        "text": "Running recall",
+        "line": "Running recall · 1s"
+      }
+    },
+    {
+      "about": "An approval, naming the tool.",
+      "progress": {
+        "phase": "waiting_for_person",
+        "since": 1762790400000,
+        "what": "approval",
+        "tool": "bash"
+      },
+      "now": 1762790430000,
+      "expect": {
+        "text": "Waiting for you: allow bash?",
+        "line": "Waiting for you: allow bash? · 30s"
+      }
+    },
+    {
+      "about": "An approval with no tool named.",
+      "progress": {
+        "phase": "waiting_for_person",
+        "since": 1762790400000,
+        "what": "approval"
+      },
+      "now": 1762790400000,
+      "expect": {
+        "text": "Waiting for you: an approval",
+        "line": "Waiting for you: an approval · 0s"
+      }
+    },
+    {
+      "about": "A question.",
+      "progress": {
+        "phase": "waiting_for_person",
+        "since": 1762790400000,
+        "what": "question"
+      },
+      "now": 1762790403000,
+      "expect": {
+        "text": "Waiting for you: a question",
+        "line": "Waiting for you: a question · 3s"
+      }
+    },
+    {
+      "about": "The plan.",
+      "progress": {
+        "phase": "waiting_for_person",
+        "since": 1762790400000,
+        "what": "plan"
+      },
+      "now": 1762790403000,
+      "expect": {
+        "text": "Waiting for you: the plan",
+        "line": "Waiting for you: the plan · 3s"
+      }
+    },
+    {
+      "about": "Subagents.",
+      "progress": {
+        "phase": "waiting_for_subagents",
+        "since": 1762790400000,
+        "running": 3
+      },
+      "now": 1762790490000,
+      "expect": {
+        "text": "Waiting for subagents: 3 running",
+        "line": "Waiting for subagents: 3 running · 1m 30s"
+      }
+    },
+    {
+      "about": "Compacting.",
+      "progress": {
+        "phase": "compacting",
+        "since": 1762790400000
+      },
+      "now": 1762790407000,
+      "expect": {
+        "text": "Compacting the conversation",
+        "line": "Compacting the conversation · 7s"
+      }
+    },
+    {
+      "about": "Retrying: `in` counts down from since.",
+      "progress": {
+        "phase": "retrying",
+        "since": 1762790400000,
+        "reason": "502 Bad Gateway",
+        "in": 10,
+        "attempt": 2
+      },
+      "now": 1762790403000,
+      "expect": {
+        "text": "Retrying in 7 s: 502 Bad Gateway",
+        "line": "Retrying in 7 s: 502 Bad Gateway · 3s"
+      }
+    },
+    {
+      "about": "Retrying: never below 0.",
+      "progress": {
+        "phase": "retrying",
+        "since": 1762790400000,
+        "reason": "overloaded",
+        "in": 2
+      },
+      "now": 1762790405000,
+      "expect": {
+        "text": "Retrying in 0 s: overloaded",
+        "line": "Retrying in 0 s: overloaded · 5s"
+      }
+    },
+    {
+      "about": "Stopped draws no elapsed time.",
+      "progress": {
+        "phase": "stopped",
+        "since": 1762790400000
+      },
+      "now": 1762790405000,
+      "expect": {
+        "text": "Stopped",
+        "line": "Stopped"
+      }
+    },
+    {
+      "about": "Stopped by the person.",
+      "progress": {
+        "phase": "stopped",
+        "since": 1762790400000,
+        "reason": "person"
+      },
+      "now": 1762790405000,
+      "expect": {
+        "text": "Stopped: by you",
+        "line": "Stopped: by you"
+      }
+    },
+    {
+      "about": "Stopped at the step limit.",
+      "progress": {
+        "phase": "stopped",
+        "since": 1762790400000,
+        "reason": "steps"
+      },
+      "now": 1762790400000,
+      "expect": {
+        "text": "Stopped: step limit",
+        "line": "Stopped: step limit"
+      }
+    },
+    {
+      "about": "Failed.",
+      "progress": {
+        "phase": "failed",
+        "since": 1762790400000,
+        "reason": "connection refused"
+      },
+      "now": 1762790405000,
+      "expect": {
+        "text": "Failed: connection refused",
+        "line": "Failed: connection refused"
+      }
+    }
+  ]
+}

tests/ui/fixtures/SOURCE

@@ -0,0 +1 @@
+LLeMbas v2.0.0-beta.8 (d06ca353b801916302f0ce0cc9ae3e5244ff184f) tests/fixtures/api_v2

tests/ui/fixtures/chat_agent.json

@@ -0,0 +1,31 @@
+{
+  "id": "chat2",
+  "kind": "agent",
+  "title": "Waybread",
+  "model": "qwen",
+  "folder_id": null,
+  "pinned": false,
+  "archived": false,
+  "temporary": false,
+  "unread": true,
+  "running": true,
+  "mode": "edit",
+  "effort": "low",
+  "device": {
+    "token_id": "token1",
+    "directory": "/home/dev/project"
+  },
+  "device_session_id": "ses_01waybread",
+  "parent_chat_id": null,
+  "agents": {
+    "min": 1,
+    "max": 4
+  },
+  "subagents": {
+    "offered": true,
+    "named_only": false,
+    "cap": 32
+  },
+  "created_at": "2026-01-02T03:04:05+00:00",
+  "updated_at": "2026-01-02T03:04:05+00:00"
+}

tests/ui/fixtures/event_ask.json

@@ -0,0 +1,162 @@
+{
+  "cards": [
+    {
+      "id": "card1",
+      "kind": "question",
+      "why": "",
+      "dismissable": false,
+      "expires_in": null,
+      "items": [
+        {
+          "index": 0,
+          "key": "q0",
+          "kind": "question",
+          "tool": "ask_user",
+          "title": "Honey or not?",
+          "header": "Taste",
+          "detail": "",
+          "reason": "",
+          "purpose": "",
+          "options": [
+            {
+              "label": "Honey",
+              "description": "",
+              "recommended": true
+            },
+            {
+              "label": "Plain",
+              "description": "As the elves have it",
+              "recommended": false
+            }
+          ],
+          "multiple": false,
+          "free_text": true,
+          "editable": false,
+          "plan": null,
+          "mode_choice": true,
+          "job": "",
+          "subagent": null
+        }
+      ]
+    },
+    {
+      "id": "card2",
+      "kind": "approval",
+      "why": "",
+      "dismissable": false,
+      "expires_in": null,
+      "items": [
+        {
+          "index": 0,
+          "key": "0",
+          "kind": "approval",
+          "tool": "bash",
+          "title": "Run a command",
+          "header": "",
+          "detail": "rm -rf build",
+          "reason": "Edit mode asks before anything that runs a command.",
+          "purpose": "Start clean",
+          "options": [],
+          "multiple": false,
+          "free_text": false,
+          "editable": true,
+          "plan": null,
+          "mode_choice": true,
+          "job": "explore (job1)",
+          "subagent": null
+        }
+      ]
+    },
+    {
+      "id": "card3",
+      "kind": "plan",
+      "why": "",
+      "dismissable": true,
+      "expires_in": null,
+      "items": [
+        {
+          "index": 0,
+          "key": "p0",
+          "kind": "plan",
+          "tool": "plan_submit",
+          "title": "Bake lembas",
+          "header": "",
+          "detail": "",
+          "reason": "",
+          "purpose": "",
+          "options": [],
+          "multiple": false,
+          "free_text": true,
+          "editable": false,
+          "plan": {
+            "version": 2,
+            "title": "Bake lembas",
+            "summary": "",
+            "findings": [],
+            "objectives": [],
+            "phases": [
+              {
+                "id": "p1",
+                "title": "Plan",
+                "status": "pending",
+                "tasks": [
+                  {
+                    "id": "t1",
+                    "text": "Gather mallorn leaves",
+                    "status": "todo",
+                    "note": ""
+                  },
+                  {
+                    "id": "t2",
+                    "text": "Bake slowly",
+                    "status": "todo",
+                    "note": ""
+                  }
+                ]
+              }
+            ],
+            "steps": [
+              "Gather mallorn leaves",
+              "Bake slowly"
+            ]
+          },
+          "mode_choice": true,
+          "job": "",
+          "subagent": null
+        }
+      ]
+    },
+    {
+      "id": "card4",
+      "kind": "approval",
+      "why": "",
+      "dismissable": false,
+      "expires_in": null,
+      "items": [
+        {
+          "index": 0,
+          "key": "0",
+          "kind": "approval",
+          "tool": "bash",
+          "title": "Run a command",
+          "header": "",
+          "detail": "git push",
+          "reason": "Edit mode asks before anything that runs a command.",
+          "purpose": "",
+          "options": [],
+          "multiple": false,
+          "free_text": false,
+          "editable": true,
+          "plan": null,
+          "mode_choice": true,
+          "job": "",
+          "subagent": {
+            "id": "s1",
+            "name": "parser",
+            "position": "reviews the parser"
+          }
+        }
+      ]
+    }
+  ]
+}

tests/ui/fixtures/event_status.json

@@ -0,0 +1,9 @@
+{
+  "text": "Running bake --slow",
+  "progress": {
+    "phase": "running_tool",
+    "since": 1767323045000,
+    "tool": "bash",
+    "what": "bake --slow"
+  }
+}

tests/ui/fixtures/event_status_idle.json

@@ -0,0 +1,4 @@
+{
+  "text": "",
+  "progress": null
+}

tests/ui/fixtures/event_status_person.json

@@ -0,0 +1,8 @@
+{
+  "text": "Waiting for you to allow bash…",
+  "progress": {
+    "phase": "waiting_for_person",
+    "since": 1767323045000,
+    "what": "question"
+  }
+}

tests/ui/fixtures/event_status_stopped.json

@@ -0,0 +1,8 @@
+{
+  "text": "",
+  "progress": {
+    "phase": "stopped",
+    "since": 1767323045000,
+    "reason": "person"
+  }
+}

tests/ui/fixtures/event_tree.json

@@ -0,0 +1,75 @@
+{
+  "tree": {
+    "min": 1,
+    "max": 4,
+    "size": 2,
+    "open_slots": 1,
+    "agents": [
+      {
+        "id": "lead",
+        "name": "lead",
+        "position": "the main agent",
+        "parent": null,
+        "state": "working",
+        "model": "qwen-27b",
+        "mode": "edit",
+        "request": [],
+        "manager": true,
+        "context": {
+          "used": 41200,
+          "window": 131072
+        },
+        "progress": {
+          "phase": "thinking",
+          "since": 1767323045000
+        },
+        "findings": 0,
+        "messages": 2
+      },
+      {
+        "id": "s2",
+        "name": "parser",
+        "position": "reviews the parser",
+        "parent": "lead",
+        "state": "working",
+        "model": "qwen-27b",
+        "mode": "manual",
+        "task": "Review src/parse.ts.",
+        "request": [
+          "write"
+        ],
+        "manager": false,
+        "context": {
+          "used": 9100,
+          "window": 131072
+        },
+        "progress": {
+          "phase": "running_tool",
+          "since": 1767323045000,
+          "tool": "read",
+          "what": "src/parse.ts"
+        },
+        "findings": 2,
+        "messages": 3
+      }
+    ],
+    "ended": [
+      {
+        "id": "s1",
+        "name": "scout",
+        "position": "maps the code",
+        "parent": "lead",
+        "state": "done",
+        "model": "qwen-27b",
+        "mode": "edit",
+        "task": "Map it.",
+        "request": [],
+        "manager": false,
+        "progress": null,
+        "findings": 1,
+        "messages": 1,
+        "ended": 1767323045000
+      }
+    ]
+  }
+}

tests/ui/fixtures/models_agent.json

@@ -0,0 +1,81 @@
+{
+  "models": [
+    {
+      "id": "qwen",
+      "name": "Qwen",
+      "vision": false,
+      "tools": true,
+      "efforts": [
+        "medium"
+      ],
+      "thinking_switch": true,
+      "levels": [
+        "off",
+        "medium"
+      ],
+      "effort": "medium",
+      "control": "switch",
+      "parallel": 2,
+      "context": 131072,
+      "image": "/api/v2/models/images/0a1b2c3d4e5f.png",
+      "default": true,
+      "subagents": {
+        "offered": true,
+        "named_only": false
+      }
+    },
+    {
+      "id": "gpt-oss",
+      "name": "gpt-oss",
+      "vision": false,
+      "tools": true,
+      "efforts": [
+        "low",
+        "medium",
+        "high"
+      ],
+      "thinking_switch": false,
+      "levels": [
+        "low",
+        "medium",
+        "high"
+      ],
+      "effort": "high",
+      "control": "slider",
+      "parallel": 1,
+      "context": 0,
+      "image": null,
+      "default": false,
+      "subagents": {
+        "offered": true,
+        "named_only": false
+      }
+    },
+    {
+      "id": "vision-small",
+      "name": "Vision",
+      "vision": true,
+      "tools": false,
+      "efforts": [],
+      "thinking_switch": false,
+      "levels": [
+        "default"
+      ],
+      "effort": "default",
+      "control": "none",
+      "parallel": null,
+      "context": 8192,
+      "image": null,
+      "default": false,
+      "subagents": {
+        "offered": true,
+        "named_only": false
+      }
+    }
+  ],
+  "subagents_cap": 32,
+  "agent_subagents": {
+    "offered": true,
+    "device_runs": true
+  }
+}

tests/ui/fixtures/subagent_findings.json

@@ -0,0 +1,10 @@
+{
+  "id": "s2",
+  "findings": [
+    {
+      "n": 1,
+      "text": "Login is POST /session.",
+      "time": 1767323045000
+    }
+  ]
+}

tests/ui/fixtures/subagent_transcript.json

@@ -0,0 +1,24 @@
+{
+  "id": "s2",
+  "state": "working",
+  "turns": [
+    {
+      "role": "user",
+      "text": "Write the routes.",
+      "machine": true
+    },
+    {
+      "role": "assistant",
+      "text": "Reading the API first.",
+      "tools": [
+        {
+          "name": "read",
+          "title": "src/api.py",
+          "isError": false
+        }
+      ],
+      "error": "",
+      "running": true
+    }
+  ]
+}

tests/ui/fixtures/tree.json

@@ -0,0 +1,112 @@
+{
+  "tree": {
+    "min": 2,
+    "max": 4,
+    "size": 3,
+    "open_slots": 0,
+    "agents": [
+      {
+        "id": "lead",
+        "name": "lead",
+        "position": "the main agent",
+        "parent": null,
+        "state": "working",
+        "model": "local/qwen",
+        "mode": "auto",
+        "request": [],
+        "manager": true,
+        "progress": {
+          "phase": "waiting_for_subagents",
+          "since": 1767323045000,
+          "running": 2
+        },
+        "findings": 0,
+        "messages": 1
+      },
+      {
+        "id": "s1",
+        "name": "area",
+        "position": "owns the API",
+        "parent": "lead",
+        "state": "working",
+        "model": "local/qwen",
+        "mode": "auto",
+        "request": [],
+        "manager": true,
+        "task": "Direct it.",
+        "progress": null,
+        "findings": 0,
+        "messages": 1
+      },
+      {
+        "id": "s2",
+        "name": "routes",
+        "position": "writes the routes",
+        "parent": "area",
+        "state": "working",
+        "model": "local/qwen",
+        "mode": "edit",
+        "request": [
+          "execute"
+        ],
+        "manager": false,
+        "task": "Write the routes.",
+        "context": {
+          "used": 9100,
+          "window": 131072
+        },
+        "progress": {
+          "phase": "running_tool",
+          "since": 1767323045000,
+          "tool": "read",
+          "what": "src/api.py"
+        },
+        "findings": 1,
+        "messages": 1
+      }
+    ],
+    "ended": [
+      {
+        "id": "s3",
+        "name": "scout",
+        "position": "maps the code",
+        "parent": "lead",
+        "state": "done",
+        "model": "local/qwen",
+        "mode": "auto",
+        "request": [],
+        "manager": false,
+        "task": "Map it.",
+        "progress": null,
+        "findings": 0,
+        "messages": 1,
+        "ended": 1767323045000
+      }
+    ]
+  },
+  "room": [
+    {
+      "n": 1,
+      "from": "routes",
+      "to": [
+        "area"
+      ],
+      "text": "Routes are in.",
+      "report": false,
+      "kind": "message",
+      "time": 1767323045000
+    },
+    {
+      "n": 2,
+      "from": "scout",
+      "to": [
+        "lead"
+      ],
+      "text": "Mapped: see map.md.",
+      "report": true,
+      "kind": "report",
+      "time": 1767323045000
+    }
+  ],
+  "live": true
+}

tests/ui/mock-tauri.js

@@ -0,0 +1,140 @@
+// A stand-in for Tauri's bridge, for the page under a headless browser (tests/ui/run.mjs). Runs
+// before the page's own scripts: `window.__TAURI__` answers the commands the page invokes from a
+// table the test hands in, and records every call, so a test can say what the page asked for.
+//
+// The table (`config`):
+//   status   what `get_status` answers (the app, the agent, the server)
+//   routes   { "GET /me": value, "POST /chats/c1/interactions/k1": value, ... } -- the server's
+//            /api/v2 as the page reaches it through `api`. A path is looked up with its query,
+//            then without it. A value `{ __error: { status, detail } }` is a refusal. Anything
+//            not in the table is a 404, as an older server would answer.
+//   streams  { "/chats/c1/messages/m1/stream": [[event, data], ...] } -- the frames a followed
+//            reply sends as soon as it is opened. More can be sent later with `__mock.frame()`.
+//   terminal [[kind, data], ...] -- what every terminal's socket sends as soon as it is opened
+//            (`control` frames as the server's JSON, `output` as text, sent base64 as Rust does).
+//            More with `__mock.shell(kind, data)`; `{ refuse: {status, detail} }` makes
+//            `terminal_open` fail.
+//
+// Nothing here talks to a network: the page's policy allows it none, and this keeps it so.
+;(function () {
+  const config = window.__mockConfig || {}
+  const listeners = new Map() // event name -> [callback]
+  const streams = new Map() // stream id -> path
+  const terminals = new Map() // terminal id -> { target, open }
+  let nextTerminal = 0
+  let nextStream = 0
+  const calls = []
+
+  function dispatch(name, payload) {
+    for (const cb of listeners.get(name) || []) cb({ event: name, payload })
+  }
+
+  function route(method, path) {
+    const routes = config.routes || {}
+    const bare = path.split("?")[0]
+    for (const key of [`${method} ${path}`, `${method} ${bare}`]) {
+      if (Object.prototype.hasOwnProperty.call(routes, key)) {
+        const value = routes[key]
+        if (value && value.__error) return Promise.reject(value.__error)
+        return Promise.resolve(JSON.parse(JSON.stringify(value)))
+      }
+    }
+    return Promise.reject({ status: 404, detail: `Not in the test's table: ${method} ${path}` })
+  }
+
+  // The shell's bytes, base64 as Rust sends them (`terminal.rs`): UTF-8 first.
+  const b64 = (text) => btoa(String.fromCharCode(...new TextEncoder().encode(text)))
+
+  function shell(kind, data, id) {
+    const ids = id ? [id] : [...terminals.keys()].filter((k) => terminals.get(k).open)
+    for (const tid of ids) {
+      if (kind === "end") terminals.get(tid).open = false
+      dispatch("terminal", { id: tid, kind, data: kind === "output" ? b64(data) : data })
+    }
+  }
+
+  function frame(path, event, data) {
+    for (const [id, p] of streams) if (p === path) dispatch("stream", { id, event, data })
+  }
+
+  async function invoke(cmd, args) {
+    calls.push({ cmd, args: args === undefined ? null : JSON.parse(JSON.stringify(args)) })
+    switch (cmd) {
+      case "get_status":
+        return JSON.parse(JSON.stringify(config.status))
+      case "api":
+        return route(args.method, args.path)
+      case "api_text":
+        return route("GET", args.path)
+      case "stream_open": {
+        const id = ++nextStream
+        streams.set(id, args.path)
+        const frames = (config.streams || {})[args.path] || []
+        setTimeout(() => {
+          for (const [event, data] of frames) dispatch("stream", { id, event, data })
+        }, 0)
+        return id
+      }
+      case "stream_close":
+        streams.delete(args.id)
+        return null
+      case "terminal_open": {
+        const t = config.terminal || []
+        if (t.refuse) return Promise.reject(t.refuse)
+        const id = ++nextTerminal
+        terminals.set(id, { target: { chat: args.chat, device: args.device, directory: args.directory }, open: true })
+        setTimeout(() => {
+          for (const [kind, data] of t) shell(kind, data, id)
+        }, 0)
+        return id
+      }
+      case "terminal_input":
+      case "terminal_resize":
+        return !!terminals.get(args.id)?.open
+      case "terminal_close": {
+        const t = terminals.get(args.id)
+        if (t?.open) setTimeout(() => shell("end", { code: 1000, reason: "", status: null, error: null }, args.id), 0)
+        return null
+      }
+      case "model_image":
+        return ""
+      case "terminal_word":
+        return "lembas"
+      default:
+        return null
+    }
+  }
+
+  const win = {
+    minimize: async () => {},
+    toggleMaximize: async () => {},
+    close: async () => {},
+    isMaximized: async () => false,
+  }
+
+  window.__TAURI__ = {
+    core: { invoke },
+    event: {
+      listen: async (name, cb) => {
+        if (!listeners.has(name)) listeners.set(name, [])
+        listeners.get(name).push(cb)
+        return () => listeners.set(name, (listeners.get(name) || []).filter((x) => x !== cb))
+      },
+    },
+    window: { getCurrentWindow: () => win },
+  }
+
+  window.__mock = {
+    calls,
+    frame,
+    emit: dispatch,
+    /** The calls of one command, or of `api` with one method and path prefix. */
+    api: (method, prefix) => calls.filter((c) => c.cmd === "api" && c.args.method === method && c.args.path.startsWith(prefix)),
+    streams: () => [...streams.values()],
+    shell,
+    /** The terminals opened so far: [{ id, target, open }]. */
+    terminals: () => [...terminals].map(([id, t]) => ({ id, ...t })),
+    /** What was typed into a terminal, joined. */
+    typed: () => calls.filter((c) => c.cmd === "terminal_input").map((c) => c.args.data).join(""),
+  }
+})()

tests/ui/run.mjs

@@ -0,0 +1,842 @@
+#!/usr/bin/env node
+// The page under a headless browser, against a mocked bridge (tests/ui/mock-tauri.js): what the
+// person sees for a given answer of the server's /api/v2, and what the page sends back.
+//
+//     node tests/ui/run.mjs [name-filter]
+//
+// Needs Playwright (`playwright` or `playwright-core`, from NODE_PATH or a node_modules beside
+// the repository) and a Chromium: Playwright's own, or the one in $CHROMIUM (/usr/bin/chromium
+// is tried when Playwright has none). Nothing is built and no server is started: the page's
+// files are handed to the browser from ui/ as they are in the repository. With $SHOTS set to a
+// directory, some tests leave a screenshot there.
+//
+// The shapes are the server's (`api/v2/serialise.py`, its fixtures in tests/fixtures/api_v2),
+// with made-up names. Where the server's fixture differs from what its code sends, the test
+// says so and follows the code.
+
+import { createRequire } from "node:module"
+import { readFileSync, existsSync } from "node:fs"
+import { dirname, join, extname } from "node:path"
+import { fileURLToPath } from "node:url"
+import assert from "node:assert/strict"
+
+const HERE = dirname(fileURLToPath(import.meta.url))
+const UI = join(HERE, "..", "..", "ui")
+const ORIGIN = "http://app.test"
+const require = createRequire(join(process.cwd(), "noop.js"))
+
+function playwright() {
+  for (const name of ["playwright", "playwright-core"]) {
+    try {
+      return require(name)
+    } catch {}
+    try {
+      return createRequire(import.meta.url)(name)
+    } catch {}
+  }
+  console.error("tests/ui: Playwright is not installed (npm install playwright-core, and put it on NODE_PATH)")
+  process.exit(2)
+}
+
+async function launch() {
+  const { chromium } = playwright()
+  const paths = [process.env.CHROMIUM, "/usr/bin/chromium", "/usr/bin/chromium-browser"].filter(Boolean)
+  if (!process.env.CHROMIUM) {
+    try {
+      return await chromium.launch()
+    } catch {}
+  }
+  for (const executablePath of paths) if (existsSync(executablePath)) return chromium.launch({ executablePath })
+  throw new Error("tests/ui: no Chromium found (set CHROMIUM)")
+}
+
+/** One of the server's /api/v2 contract fixtures, as copied by scripts/fetch-fixtures.sh. */
+const fixture = (name) => JSON.parse(readFileSync(join(HERE, "fixtures", `${name}.json`), "utf8"))
+
+const TYPES = { ".html": "text/html", ".js": "text/javascript", ".css": "text/css", ".svg": "image/svg+xml", ".mjs": "text/javascript", ".json": "application/json" }
+
+// --- The server, as the tests' table -----------------------------------------------------------
+
+const STAMP = "2026-01-02T03:04:05+00:00"
+
+function status(over = {}) {
+  return {
+    version: "1.0.0-test",
+    server: "https://ai.example",
+    discovery: { name: "Example", app: "https://ai.example/api/v2", version: "2.0.0" },
+    refused: null,
+    checked: true,
+    account: { name: "Ada Example", email: "ada@example.org" },
+    signing_in: false,
+    linked: true,
+    agent: { state: "running", detail: null },
+    agent_wanted: true,
+    update: null,
+    remote: { enabled: true, roots: ["/home/dev/project"] },
+    theme: "moria",
+    host: "workbench",
+    notify: false,
+    ...over,
+  }
+}
+
+const SUB = (offered, named_only = false) => ({ offered, named_only })
+
+/** The models as `/api/v2/models` lists them; `subagents` per model as the server works it out. */
+function models({ qwen = SUB(true), solo = SUB(true, true), plain = SUB(false) } = {}) {
+  const base = { vision: false, tools: true, efforts: ["low", "medium", "high"], thinking_switch: false, levels: ["low", "medium", "high"], effort: "medium", control: "slider", context: 131072, image: null }
+  return {
+    models: [
+      { ...base, id: "qwen", name: "Qwen", parallel: 2, default: true, subagents: qwen },
+      { ...base, id: "solo", name: "Solo", parallel: 1, default: false, subagents: solo },
+      { ...base, id: "plain", name: "Plain", parallel: 1, default: false, subagents: plain },
+    ],
+    subagents_cap: 8,
+  }
+}
+
+function chatRow(over = {}) {
+  return {
+    id: "chat1",
+    kind: "chat",
+    title: "Waybread",
+    model: "qwen",
+    folder_id: null,
+    pinned: false,
+    archived: false,
+    temporary: false,
+    unread: false,
+    running: false,
+    mode: "",
+    effort: "medium",
+    device: null,
+    device_session_id: null,
+    parent_chat_id: null,
+    agents: { min: 1, max: 4 },
+    subagents: { offered: true, named_only: false, cap: 8 },
+    created_at: STAMP,
+    updated_at: STAMP,
+    ...over,
+  }
+}
+
+function agentChat(over = {}) {
+  return chatRow({
+    id: "chat2",
+    kind: "agent",
+    title: "Tidy the build",
+    running: true,
+    mode: "edit",
+    device: { token_id: "token1", directory: "/home/dev/project" },
+    device_session_id: "ses_01",
+    // An agent chat's team runs on its device: the server offers no setting (team_run.may_have_team).
+    subagents: { offered: false, named_only: false, cap: 8 },
+    ...over,
+  })
+}
+
+const userMessage = (id, text) => ({ id, role: "user", created_at: STAMP, complete: true, stopped: false, error: null, queued: false, machine: false, origin: "", model: null, attachments: [], text })
+
+/**
+ * A device's bash approval as `services/device_link.py:_permission` builds it and
+ * `api/v2/serialise.py:interruption` sends it: the title is the fixed "Run a command", the command
+ * is in `detail`, and it is `editable`. (The server's fixture `event_ask.json` puts the command in
+ * the title; its code does not.)
+ */
+function deviceBashCard(over = {}) {
+  return {
+    id: "card7",
+    kind: "approval",
+    why: "",
+    dismissable: false,
+    expires_in: 1800,
+    items: [
+      {
+        index: 0,
+        key: "0",
+        kind: "approval",
+        tool: "bash",
+        title: "Run a command",
+        header: "",
+        detail: "ls -la build",
+        reason: "Edit mode asks before anything that runs a command.",
+        purpose: "See what is in the build directory",
+        options: [],
+        multiple: false,
+        free_text: true,
+        editable: true,
+        plan: null,
+        mode_choice: false,
+        job: "",
+      },
+    ],
+    ...over,
+  }
+}
+
+function baseRoutes(over = {}) {
+  return {
+    "GET /me": { id: "u1", name: "Ada Example", email: "ada@example.org", theme: "moria", language: "en", permissions: {} },
+    "GET /models": models(),
+    "GET /devices": { devices: [{ id: "token1", name: "workbench", host: "workbench", platform: "linux", terminal: true }] },
+    "GET /chats": { chats: [chatRow(), agentChat()], more: false },
+    "GET /schedules": { schedules: [] },
+    ...over,
+  }
+}
+
+// --- The browser --------------------------------------------------------------------------------
+
+let browser
+
+async function page(config, { width = 1280, height = 860 } = {}) {
+  const context = await browser.newContext({ viewport: { width, height } })
+  const p = await context.newPage()
+  const errors = []
+  p.on("pageerror", (e) => errors.push(String(e)))
+  await context.route(`${ORIGIN}/**`, (route) => {
+    const path = new URL(route.request().url()).pathname
+    const file = join(UI, path === "/" ? "index.html" : decodeURIComponent(path))
+    if (!file.startsWith(UI) || !existsSync(file)) return route.fulfill({ status: 404, body: "" })
+    return route.fulfill({ status: 200, contentType: TYPES[extname(file)] || "application/octet-stream", body: readFileSync(file) })
+  })
+  await p.addInitScript((c) => {
+    window.__mockConfig = c
+  }, config)
+  await p.addInitScript({ path: join(HERE, "mock-tauri.js") })
+  await p.goto(`${ORIGIN}/index.html`)
+  await p.waitForSelector("#shell:not([hidden])")
+  p.errors = errors
+  return p
+}
+
+async function openChatRow(p, id) {
+  await p.click(`.chatrow[data-id="${id}"]`)
+  await p.waitForSelector(`#messages .msg`)
+}
+
+// --- The tests ----------------------------------------------------------------------------------
+
+const tests = []
+const test = (name, fn) => tests.push({ name, fn })
+
+const STREAM2 = "/chats/chat2/messages/m2/stream"
+
+function agentChatOpen(frames, over = {}) {
+  return {
+    status: status(),
+    routes: baseRoutes({
+      "GET /chats/chat2": { chat: agentChat(), messages: [userMessage("m1", "What is in build?")], more: false, live: "m2" },
+      "GET /chats/chat1": { chat: chatRow(), messages: [userMessage("m5", "How is lembas made?")], more: false, live: null },
+      "POST /chats/chat2/interactions/card8": { answered: true, remembered: false, not_remembered: [] },
+      "GET /models?chat=chat2": models(),
+      "POST /chats/chat2/interactions/card7": { answered: true, remembered: false, not_remembered: [] },
+      ...over,
+    }),
+    streams: { [STREAM2]: frames },
+  }
+}
+
+test("A1: an agent chat's device bash approval shows as a card, the command editable, and Allow answers it", async () => {
+  const card = deviceBashCard()
+  const p = await page(
+    agentChatOpen([
+      ["steps", { steps: [], tail: { thinking: false, text: "", label: "" } }],
+      ["status", { text: "Waiting for you to allow bash…" }],
+      ["ask", { cards: [card] }],
+    ]),
+  )
+  await openChatRow(p, "chat2")
+  await p.waitForSelector("#cards:not([hidden]) .ask")
+  assert.equal(await p.textContent(".msg.is-live .msg__status"), "Waiting for you to allow bash…")
+  const cardBox = await p.locator("#cards .ask").boundingBox()
+  const view = p.viewportSize()
+  assert.ok(cardBox && cardBox.height > 40, "the card has a size")
+  assert.ok(cardBox.y >= 0 && cardBox.y + 20 <= view.height, `the card is on the screen (y=${cardBox.y})`)
+  assert.ok(await p.locator("#cards .ask").isVisible(), "the card is visible")
+  // The command is the item's `detail`, in the box to correct it; the title is not the command.
+  assert.equal(await p.inputValue("#cards .ask__edit"), "ls -la build")
+  const text = await p.textContent("#cards .ask")
+  assert.match(text, /bash/)
+  assert.match(text, /See what is in the build directory/, "the model's purpose is said")
+  assert.match(text, /Edit mode asks before anything that runs a command\./, "the server's reason is said")
+  if (process.env.SHOTS) await p.screenshot({ path: join(process.env.SHOTS, "a1-card.png") })
+  await p.click("#cards .ask__actions .btn--primary")
+  await p.waitForFunction(() => window.__mock.api("POST", "/chats/chat2/interactions/").length === 1)
+  const [post] = await p.evaluate(() => window.__mock.api("POST", "/chats/chat2/interactions/"))
+  assert.equal(post.args.path, "/chats/chat2/interactions/card7")
+  assert.deepEqual(post.args.body, { verdict: "allow", choices: {}, texts: {} }, "an untouched command is not sent as an edit")
+  await p.waitForSelector("#cards", { state: "hidden" })
+  assert.deepEqual(p.errors, [])
+})
+
+test("A1: a corrected command is sent as the item's text; Deny sends deny", async () => {
+  const p = await page(agentChatOpen([["ask", { cards: [deviceBashCard()] }]]))
+  await openChatRow(p, "chat2")
+  await p.waitForSelector("#cards .ask__edit")
+  await p.fill("#cards .ask__edit", "ls -la build/out")
+  await p.click("#cards .ask__actions .btn--primary")
+  await p.waitForFunction(() => window.__mock.api("POST", "/chats/chat2/interactions/").length === 1)
+  const [post] = await p.evaluate(() => window.__mock.api("POST", "/chats/chat2/interactions/"))
+  assert.deepEqual(post.args.body, { verdict: "allow", choices: {}, texts: { 0: "ls -la build/out" } })
+  // The next one, refused.
+  await p.evaluate(([path, c]) => window.__mock.frame(path, "ask", { cards: [c] }), [STREAM2, deviceBashCard({ id: "card8" })])
+  await p.waitForSelector("#cards .ask__edit")
+  await p.click("#cards .ask__actions .btn >> text=Deny")
+  await p.waitForFunction(() => window.__mock.api("POST", "/chats/chat2/interactions/card8").length === 1)
+  const [deny] = await p.evaluate(() => window.__mock.api("POST", "/chats/chat2/interactions/card8"))
+  assert.deepEqual(deny.args.body, { verdict: "deny", choices: {}, texts: {} })
+  assert.deepEqual(p.errors, [])
+})
+
+test("A1: a card that arrives while the person is typing keeps what they typed and is not drawn again by an unchanged frame", async () => {
+  const card = deviceBashCard()
+  const p = await page(agentChatOpen([["ask", { cards: [card] }]]))
+  await openChatRow(p, "chat2")
+  await p.waitForSelector("#cards .ask__edit")
+  await p.fill("#cards .ask__edit", "ls build")
+  // The server sends the frame again when anything in it changes -- here only the countdown.
+  await p.evaluate(([path, c]) => window.__mock.frame(path, "ask", { cards: [{ ...c, expires_in: 1790 }] }), [STREAM2, card])
+  await p.evaluate(([path]) => window.__mock.frame(path, "status", { text: "Waiting for you to allow bash…" }), [STREAM2])
+  assert.equal(await p.inputValue("#cards .ask__edit"), "ls build")
+  assert.equal(await p.locator("#cards .ask").count(), 1)
+  // A second card joins the first without wiping it; the answered one goes.
+  await p.evaluate(([path, c]) => window.__mock.frame(path, "ask", { cards: [c, { ...c, id: "card8", items: [{ ...c.items[0], detail: "make clean" }] }] }), [STREAM2, card])
+  assert.equal(await p.locator("#cards .ask").count(), 2)
+  assert.equal(await p.inputValue("#cards .ask >> nth=0 >> .ask__edit"), "ls build")
+  await p.evaluate(([path]) => window.__mock.frame(path, "ask", { cards: [] }), [STREAM2])
+  await p.waitForSelector("#cards", { state: "hidden" })
+  assert.deepEqual(p.errors, [])
+})
+
+test("A1: reopening a chat whose reply waits shows the card from the stream's first pass", async () => {
+  const p = await page(agentChatOpen([["status", { text: "Waiting for you to allow bash…" }], ["ask", { cards: [deviceBashCard()] }]]))
+  await openChatRow(p, "chat2")
+  await p.waitForSelector("#cards .ask")
+  // Away to another chat and back: the stream is opened again and its first pass carries the card.
+  await p.click('.chatrow[data-id="chat1"]').catch(() => {})
+  await p.waitForFunction(() => !document.querySelector("#cards .ask"))
+  await openChatRow(p, "chat2")
+  await p.waitForSelector("#cards:not([hidden]) .ask")
+  assert.equal(await p.evaluate(() => window.__mock.calls.filter((c) => c.cmd === "stream_open" && c.args.path.includes("/messages/")).length), 2)
+  assert.deepEqual(p.errors, [])
+})
+
+test("A1: a card at 390 px wide is on the screen and its buttons can be reached", async () => {
+  const p = await page(agentChatOpen([["ask", { cards: [deviceBashCard()] }]]), { width: 390, height: 844 })
+  await p.evaluate(() => document.querySelector('.chatrow[data-id="chat2"]').click())
+  await p.waitForSelector("#cards:not([hidden]) .ask")
+  const allow = await p.locator("#cards .ask__actions .btn--primary").boundingBox()
+  assert.ok(allow && allow.y + allow.height <= 844 && allow.x >= 0 && allow.x + allow.width <= 390, `Allow is on the screen: ${JSON.stringify(allow)}`)
+  assert.ok(await p.evaluate(() => document.documentElement.scrollWidth <= 390), "nothing scrolls sideways")
+  assert.deepEqual(p.errors, [])
+})
+
+// --- A5: the Subagents setting ----------------------------------------------------------------
+//
+// The server decides (`services/team_run.py:offered`, `services/subagents.py:view`): subagents
+// enabled on the instance, the person's permission, a model that calls tools, and some model that
+// could serve one -- the chat's own serving two requests at once, or another off its one-model
+// server (`named_only`: then the setting says a subagent needs a model named in the chat). It says
+// so per model in `/models` (a new chat) and per chat in `/chats/{id}` (`subagents`). The web page
+// reads the same function (`api/pages.py:_team_context`). The app shows what it is told.
+
+async function newChatOn(p, kind, modelName) {
+  await p.click("#new-chat-btn")
+  await p.click(`.kind__tab >> text=${kind}`)
+  if (modelName) {
+    await p.click(".cchip--model")
+    await p.locator(".menu .menu__item", { hasText: modelName }).click()
+  }
+}
+
+async function subagentsChip(p) {
+  // The chip comes after the others (an open chat's waits for the server's word).
+  await p.waitForTimeout(150)
+  return p.locator("#composer-chips .cchip--subagents")
+}
+
+async function openSlider(p) {
+  await p.click("#composer-chips .cchip--subagents")
+  await p.waitForSelector(".range2")
+  return {
+    min: await p.getAttribute(".range2__input--min", "max"),
+    text: await p.textContent(".pop__body"),
+  }
+}
+
+const OWN = "This model serves one request at a time: subagents need another model, on another server, named in the chat."
+
+function a5(routes = {}) {
+  return { status: status(), routes: baseRoutes(routes), streams: {} }
+}
+
+test("A5: a new chat on a model serving two at once offers Subagents, Min and Max to the server's cap", async () => {
+  const p = await page(a5())
+  await newChatOn(p, "Chat")
+  const chip = await subagentsChip(p)
+  assert.equal(await chip.count(), 1, "the chip is there")
+  const s = await openSlider(p)
+  assert.equal(s.min, "8", "up to the server's cap")
+  assert.ok(!s.text.includes(OWN), "no one-request note")
+  assert.deepEqual(p.errors, [])
+})
+
+test("A5: a new chat on a one-request model the server still offers: Subagents with the note", async () => {
+  const p = await page(a5())
+  await newChatOn(p, "Chat", "Solo")
+  const chip = await subagentsChip(p)
+  assert.equal(await chip.count(), 1)
+  const s = await openSlider(p)
+  assert.ok(s.text.includes(OWN), "the one-request note")
+  assert.deepEqual(p.errors, [])
+})
+
+test("A5: subagents off on the instance (offered false for every model): no Subagents", async () => {
+  const off = SUB(false)
+  const p = await page(a5({ "GET /models": models({ qwen: off, solo: off, plain: off }) }))
+  await newChatOn(p, "Chat")
+  assert.equal(await (await subagentsChip(p)).count(), 0)
+  await newChatOn(p, "Chat", "Solo")
+  assert.equal(await (await subagentsChip(p)).count(), 0)
+  assert.deepEqual(p.errors, [])
+})
+
+test("A5: subagents switched on while the app stays open: the next new chat offers them", async () => {
+  const off = SUB(false)
+  const p = await page(a5({ "GET /models": models({ qwen: off, solo: off, plain: off }) }))
+  await newChatOn(p, "Chat")
+  assert.equal(await (await subagentsChip(p)).count(), 0, "off at sign-in")
+  // The administrator switches subagents on; the app read /models at sign-in.
+  await p.evaluate((m) => (window.__mockConfig.routes["GET /models"] = m), models())
+  await p.waitForTimeout(5200)
+  await p.click("#new-chat-btn")
+  await p.waitForSelector("#composer-chips .cchip--subagents")
+  assert.deepEqual(p.errors, [])
+})
+
+test("A5: an open chat follows its own `subagents`: offered, offered with the note, not offered", async () => {
+  for (const [said, chip, note] of [
+    [{ offered: true, named_only: false, cap: 8 }, 1, false],
+    [{ offered: true, named_only: true, cap: 8 }, 1, true],
+    [{ offered: false, named_only: false, cap: 8 }, 0, false],
+  ]) {
+    const p = await page(
+      a5({
+        "GET /chats/chat1": { chat: chatRow({ subagents: said }), messages: [userMessage("m5", "How is lembas made?")], more: false, live: null },
+        "GET /models?chat=chat1": models(),
+      }),
+    )
+    await openChatRow(p, "chat1")
+    const c = await subagentsChip(p)
+    assert.equal(await c.count(), chip, `chip for ${JSON.stringify(said)}`)
+    if (chip) assert.equal((await openSlider(p)).text.includes(OWN), note)
+    assert.deepEqual(p.errors, [])
+    await p.context().close()
+  }
+})
+
+// --- B3: the Terminal panel -------------------------------------------------------------------
+//
+// The server's two sockets for an app (`api/v2/terminal.py`): an agent chat's
+// (`/chats/{chat}/terminal/ws`) and one before the chat exists, on the device and folder chosen
+// (`/devices/{device}/terminal/ws?directory=`), which the chat's first message takes over.
+// `GET /devices` says per device whether it opens a terminal. The socket is Rust's: the page asks
+// `terminal_open` and gets the frames as `terminal` events (mock-tauri.js plays the server).
+
+const READY = ["control", { t: "ready", label: "workbench", dir: "/home/dev/project", cols: 80, rows: 24, shared: false, osc: false }]
+
+function b3(over = {}, terminal = [READY, ["output", "dev@workbench:~/project$ "]]) {
+  return {
+    status: status(),
+    routes: baseRoutes({
+      "GET /chats/chat2": { chat: agentChat({ running: false }), messages: [userMessage("m1", "What is in build?")], more: false, live: null },
+      "GET /models?chat=chat2": models(),
+      "POST /chats": { chat: agentChat({ running: false }), message: userMessage("m9", "Build it"), reply: null },
+      ...over,
+    }),
+    streams: {},
+    terminal,
+  }
+}
+
+async function screenText(p) {
+  return p.evaluate(() => document.querySelector("#terminal .xterm-rows")?.textContent || "")
+}
+
+test("B3: a new agent chat offers the Terminal once its folder is chosen, and opens a shell there before the first message", async () => {
+  const p = await page(b3())
+  await p.click("#new-chat-btn")
+  await p.click(".kind__tab >> text=Agent")
+  await p.waitForSelector("#composer-chips .cchip--terminal")
+  assert.equal(await p.evaluate(() => window.__mock.terminals().length), 0, "nothing is opened until asked")
+  await p.click("#composer-chips .cchip--terminal")
+  await p.waitForSelector("#terminal:not([hidden]) .xterm")
+  await p.waitForFunction(() => window.__mock.terminals().length === 1)
+  const [shell] = await p.evaluate(() => window.__mock.terminals())
+  assert.deepEqual(shell.target, { chat: null, device: "token1", directory: "/home/dev/project" })
+  await p.waitForFunction(() => document.querySelector("#terminal .xterm-rows")?.textContent.includes("dev@workbench"))
+  assert.equal(await p.textContent("#terminal-message"), "Connected.")
+  assert.equal(await p.textContent("#terminal-title"), "workbench")
+  assert.equal(await p.getAttribute("#composer-chips .cchip--terminal", "aria-pressed"), "true")
+  // Keys go up through the bridge, never a socket of the page's own.
+  await p.click("#terminal .xterm")
+  await p.keyboard.type("ls")
+  await p.keyboard.press("Enter")
+  await p.waitForFunction(() => window.__mock.typed().includes("ls\r"))
+  assert.equal(await p.evaluate(() => window.__mock.calls.filter((c) => c.cmd === "terminal_open").length), 1)
+  if (process.env.SHOTS) await p.screenshot({ path: join(process.env.SHOTS, "b3-new-1280.png") })
+  // The first message makes the chat, which takes the shell over: the same socket goes on.
+  await p.click("#composer-text")
+  await p.fill("#composer-text", "Build it")
+  await p.click("#send")
+  await p.waitForFunction(() => window.__mock.api("POST", "/chats").length === 1)
+  const [made] = await p.evaluate(() => window.__mock.api("POST", "/chats"))
+  assert.equal(made.args.body.device, "token1")
+  assert.equal(made.args.body.directory, "/home/dev/project")
+  await p.waitForSelector('#chat-tools [data-terminal-toggle][aria-pressed="true"]')
+  assert.ok(await p.isVisible("#terminal"), "the panel stays open")
+  const shells = await p.evaluate(() => window.__mock.terminals())
+  assert.equal(shells.length, 1, "no second shell: the chat took the first one over")
+  assert.ok(shells[0].open, "and its socket is still open")
+  assert.deepEqual(p.errors, [])
+})
+
+test("B3: an open agent chat opens its own shell from the header; Send to chat fills the composer and sends nothing", async () => {
+  const p = await page(b3())
+  await openChatRow(p, "chat2")
+  await p.click('#chat-tools [data-terminal-toggle]')
+  await p.waitForFunction(() => window.__mock.terminals().length === 1)
+  assert.deepEqual((await p.evaluate(() => window.__mock.terminals()))[0].target, { chat: "chat2", device: null, directory: null })
+  await p.evaluate(() => window.__mock.shell("output", "\r\ntotal 4\r\n-rw-r--r-- 1 dev dev 12 build.log\r\n"))
+  await p.waitForFunction(() => document.querySelector("#terminal .xterm-rows")?.textContent.includes("build.log"))
+  await p.fill("#composer-text", "Look at this:")
+  await p.click("#terminal-send")
+  const text = await p.inputValue("#composer-text")
+  assert.ok(text.startsWith("Look at this:\n\n```\n"), `after what was typed, as a block: ${JSON.stringify(text.slice(0, 40))}`)
+  assert.match(text, /build\.log/)
+  assert.equal(await p.evaluate(() => window.__mock.api("POST", "/chats/chat2/messages").length), 0, "nothing is sent")
+  // Closing the panel keeps the shell; opening it again does not open another.
+  await p.click("#terminal-close")
+  assert.ok(await p.isHidden("#terminal"))
+  await p.click('#chat-tools [data-terminal-toggle]')
+  await p.waitForSelector("#terminal:not([hidden])")
+  assert.equal(await p.evaluate(() => window.__mock.terminals().length), 1)
+  // The server ends it: said, and no new shell in its place.
+  await p.evaluate(() => window.__mock.shell("control", { t: "closed", message: "The shell exited." }))
+  await p.waitForFunction(() => document.querySelector("#terminal-message").textContent === "The shell exited.")
+  assert.equal(await p.evaluate(() => window.__mock.terminals().length), 1)
+  assert.deepEqual(p.errors, [])
+})
+
+test("B3: the terminal and the activity panel take turns in the panel column", async () => {
+  const p = await page(b3({ "GET /chats/chat2/activity": { activity: {} } }))
+  await openChatRow(p, "chat2")
+  await p.click('#chat-tools [data-terminal-toggle]')
+  await p.waitForSelector("#terminal:not([hidden])")
+  assert.equal(await p.getAttribute("#shell", "data-terminal"), "true")
+  await p.click("#activity-open")
+  await p.waitForSelector("#activity:not([hidden])")
+  assert.ok(await p.isHidden("#terminal"), "the terminal gave way")
+  assert.equal(await p.getAttribute("#shell", "data-terminal"), "false")
+  await p.click('#chat-tools [data-terminal-toggle]')
+  await p.waitForSelector("#terminal:not([hidden])")
+  assert.ok(await p.isHidden("#activity"), "and the activity panel to it")
+  assert.equal(await p.evaluate(() => window.__mock.terminals().length), 1, "the shell lived on behind it")
+})
+
+test("B3: no Terminal where the device opens none, for a chat on the server, or before a folder", async () => {
+  const off = { "GET /devices": { devices: [{ id: "token1", name: "workbench", host: "workbench", platform: "linux", terminal: false }] } }
+  let p = await page(b3(off))
+  await p.click("#new-chat-btn")
+  await p.click(".kind__tab >> text=Agent")
+  await p.waitForSelector("#composer-chips .cchip")
+  assert.equal(await p.locator("#composer-chips .cchip--terminal").count(), 0, "the device opens no terminal")
+  await openChatRow(p, "chat2")
+  assert.equal(await p.locator("#chat-tools [data-terminal-toggle]").count(), 0)
+  await p.context().close()
+  p = await page(b3({}))
+  await p.click("#new-chat-btn")
+  await p.click(".kind__tab >> text=Chat")
+  await p.waitForSelector("#composer-chips .cchip")
+  assert.equal(await p.locator("#composer-chips .cchip--terminal").count(), 0, "a chat on the server has no shell")
+  await p.context().close()
+  p = await page({ ...b3({}), status: status({ remote: { enabled: true, roots: [] } }) })
+  await p.click("#new-chat-btn")
+  await p.click(".kind__tab >> text=Agent")
+  await p.waitForSelector("#composer-chips .cchip")
+  assert.equal(await p.locator("#composer-chips .cchip--terminal").count(), 0, "no folder yet")
+  assert.deepEqual(p.errors, [])
+})
+
+test("B3: a refused sign-in and an older server are said in words", async () => {
+  let p = await page(b3({}, [["end", { code: 4401, reason: "not allowed", status: null, error: null }]]))
+  await openChatRow(p, "chat2")
+  await p.click('#chat-tools [data-terminal-toggle]')
+  await p.waitForFunction(() => /Sign in again/.test(document.querySelector("#terminal-message").textContent))
+  assert.ok(await p.evaluate(() => document.querySelector("#terminal-message").classList.contains("is-error")))
+  await p.context().close()
+  p = await page(b3({}, [["end", { code: null, reason: "", status: 404, error: null }]]))
+  await openChatRow(p, "chat2")
+  await p.click('#chat-tools [data-terminal-toggle]')
+  await p.waitForFunction(() => /needs a newer LLeMbas server/.test(document.querySelector("#terminal-message").textContent))
+  assert.deepEqual(p.errors, [])
+})
+
+test("B3: at 390 px the terminal is a sheet over the chat, inside the window", async () => {
+  const p = await page(b3(), { width: 390, height: 844 })
+  await p.evaluate(() => document.querySelector('.chatrow[data-id="chat2"]').click())
+  await p.waitForSelector('#chat-tools [data-terminal-toggle]')
+  await p.click('#chat-tools [data-terminal-toggle]')
+  await p.waitForFunction(() => document.querySelector("#terminal .xterm-rows")?.textContent.includes("dev@workbench"))
+  const box = await p.locator("#terminal").boundingBox()
+  assert.ok(box.x >= 0 && box.x + box.width <= 390 && box.y + box.height <= 844, `inside: ${JSON.stringify(box)}`)
+  const send = await p.locator("#terminal-send").boundingBox()
+  assert.ok(send.x + send.width <= 390 && send.y + send.height <= 844, `Send is reachable: ${JSON.stringify(send)}`)
+  const close = await p.locator("#terminal-close").boundingBox()
+  assert.ok(close.x + close.width <= 390, "Close is reachable")
+  assert.ok(await p.evaluate(() => document.documentElement.scrollWidth <= 390), "nothing scrolls sideways")
+  if (process.env.SHOTS) await p.screenshot({ path: join(process.env.SHOTS, "b3-chat-390.png") })
+  // Send to chat: the sheet steps aside, and the message box has the block.
+  await p.click("#terminal-send")
+  assert.ok(await p.isHidden("#terminal"))
+  assert.match(await p.inputValue("#composer-text"), /dev@workbench/)
+  assert.deepEqual(p.errors, [])
+})
+
+test("B3: at 1280 px the terminal sits beside the chat, which keeps its column", async () => {
+  const p = await page(b3())
+  await openChatRow(p, "chat2")
+  await p.click('#chat-tools [data-terminal-toggle]')
+  await p.waitForFunction(() => document.querySelector("#terminal .xterm-rows")?.textContent.includes("dev@workbench"))
+  const term = await p.locator("#terminal").boundingBox()
+  const main = await p.locator("#main").boundingBox()
+  assert.ok(term.x >= main.x + main.width - 1, "beside, not over")
+  assert.ok(main.width >= 400, `the chat keeps room: ${main.width}`)
+  assert.ok(term.width >= 380, `the terminal has its width: ${term.width}`)
+  if (process.env.SHOTS) await p.screenshot({ path: join(process.env.SHOTS, "b3-chat-1280.png") })
+  assert.deepEqual(p.errors, [])
+})
+
+// --- B1: the progress line ---------------------------------------------------------------------
+//
+// Harness spec 3.4.0 (harness/progress.md in the LLeMbas CLI): one `progress` object, drawn by
+// every client as `{{text}} · {{elapsed}}`. tests/ui/conformance/progress.json is a copy of the
+// spec's conformance/progress.json (3.4.0); every case must come out word for word.
+
+test("B1: every conformance case of the spec draws the same text and line", async () => {
+  const { cases } = JSON.parse(readFileSync(join(HERE, "conformance", "progress.json"), "utf8"))
+  const p = await page(b3())
+  const got = await p.evaluate(async (cs) => {
+    const m = await import("/js/progress.js")
+    return cs.map((c) => ({ text: m.progressText(c.progress, c.now), line: m.progressLine(c.progress, c.now) }))
+  }, cases)
+  cases.forEach((c, i) => assert.deepEqual(got[i], c.expect, c.about))
+  assert.deepEqual(p.errors, [])
+})
+
+test("B1: a reply being written draws the stream's progress, its clock moving, and the status text without one", async () => {
+  const p = await page(agentChatOpen([["status", { text: "Writing…" }]]))
+  await openChatRow(p, "chat2")
+  await p.waitForSelector(".msg.is-live .msg__status")
+  // An older server sends no progress at all: its status text, as before.
+  await p.evaluate((path) => window.__mock.frame(path, "status", { text: "Running bake --slow" }), STREAM2)
+  assert.equal(await p.textContent(".msg.is-live .msg__status"), "Running bake --slow")
+  // The server's status frame (event_status.json), its clock set 65 s back from now.
+  const status = fixture("event_status")
+  await p.evaluate(([path, f]) => {
+    window.__mock.frame(path, "status", { ...f, progress: { ...f.progress, since: Date.now() - 65000 } })
+  }, [STREAM2, status])
+  const line = () => p.textContent(".msg.is-live .msg__status")
+  assert.equal(await line(), "Running bash: bake --slow · 1m 05s")
+  assert.equal(await p.getAttribute(".msg.is-live .msg__status", "data-phase"), "running_tool")
+  await p.waitForFunction(() => document.querySelector(".msg.is-live .msg__status").textContent.endsWith("1m 06s"), null, { timeout: 2500 })
+  // A data-only frame with progress (the ask frame carries it too): the line follows.
+  await p.evaluate((path) => window.__mock.frame(path, "ask", { cards: [], progress: { phase: "waiting_for_person", since: Date.now(), what: "approval", tool: "bash" } }), STREAM2)
+  assert.equal(await line(), "Waiting for you: allow bash? · 0s")
+  // Nothing running (event_status_idle.json: no text, `progress: null`): the plain words again.
+  await p.evaluate(([path, f]) => window.__mock.frame(path, "status", f), [STREAM2, fixture("event_status_idle")])
+  assert.equal(await line(), "Writing…")
+  // Waiting for a person's answer (event_status_person.json): the phase, not the text.
+  await p.evaluate(([path, f]) => window.__mock.frame(path, "status", { ...f, progress: { ...f.progress, since: Date.now() } }), [STREAM2, fixture("event_status_person")])
+  assert.equal(await line(), "Waiting for you: a question · 0s")
+  // Stopped and failed have no clock.
+  await p.evaluate(([path, f]) => window.__mock.frame(path, "status", f), [STREAM2, fixture("event_status_stopped")])
+  assert.equal(await line(), "Stopped: by you")
+  assert.deepEqual(p.errors, [])
+})
+
+// --- C4: subagents (spec 3.4.0) ---------------------------------------------------------------
+//
+// The server's /api/v2 (its fixtures, copied into tests/ui/fixtures): GET /chats/{chat}/tree ->
+// {tree, room, live}, the stream's `tree` frame, a subagent's transcript and findings, POST
+// .../stop; an agent chat's device away -> the kept tree (`live` false) and 409 for the rest;
+// /models?kind=agent&device= for a new agent chat's Subagents; `subagent` on an approval item.
+
+const ACTIVITY = { activity: { version: "v1", summary: { active: 0 }, agents: [], context: null, outputs: [], team: null, plan: null } }
+
+function c4(over = {}) {
+  return b3({
+    "GET /chats/chat2/activity": ACTIVITY,
+    "GET /chats/chat2/tree": fixture("tree"),
+    "GET /chats/chat2/subagents/s2/transcript": fixture("subagent_transcript"),
+    "GET /chats/chat2/subagents/s2/findings": fixture("subagent_findings"),
+    "POST /chats/chat2/subagents/s2/stop": { stopped: ["s2"] },
+    ...over,
+  })
+}
+
+async function openTree(p) {
+  await openChatRow(p, "chat2")
+  await p.click("#activity-open")
+  await p.waitForSelector("#activity .subagents .subagent")
+}
+
+const rows = (p) => p.evaluate(() => [...document.querySelectorAll("#activity .subagents__list > .subagent")].map((r) => [r.dataset.id, r.style.getPropertyValue("--depth")]))
+
+test("C4: the activity panel draws the server's tree: nested by parent, slots, the ended ones apart", async () => {
+  const p = await page(c4())
+  await openTree(p)
+  assert.deepEqual(await rows(p), [["lead", "0"], ["s1", "1"], ["s2", "2"]])
+  const routes = await p.textContent('#activity .subagent[data-id="s2"]')
+  assert.match(routes, /routes/)
+  assert.match(routes, /writes the routes/)
+  assert.match(routes, /Running read: src\/api\.py/)
+  assert.match(routes, /6%/, "context: floor(9100 × 100 / 131072)")
+  assert.match(routes, /edit \+ execute/)
+  assert.match(routes, /messages: 1/)
+  assert.match(routes, /routes → area: Routes are in\./, "the last room message to or from it")
+  assert.match(await p.textContent('#activity .subagent[data-id="lead"]'), /the main agent/)
+  assert.match(await p.textContent("#activity .subagents"), /3 of 4/)
+  assert.match(await p.textContent("#activity .subagents"), /They run on the computer this chat runs on\./)
+  // The ended one, folded apart, with how it ended, and no stop square.
+  await p.click("#activity .subagents__ended > summary")
+  const scout = await p.textContent('#activity .subagents__ended .subagent[data-id="s3"]')
+  assert.match(scout, /done/)
+  assert.equal(await p.locator('#activity .subagent[data-id="s3"] button.agent__stop').count(), 0)
+  assert.equal(await p.locator('#activity .subagent[data-id="lead"] button.agent__stop').count(), 0, "the main agent is not stopped from here")
+  if (process.env.SHOTS) await p.screenshot({ path: join(process.env.SHOTS, "c4-tree-1280.png") })
+  assert.deepEqual(p.errors, [])
+})
+
+test("C4: a subagent's conversation and findings open read-only; Stop asks, then stops it", async () => {
+  const p = await page(c4())
+  await openTree(p)
+  await p.click('#activity .subagent[data-id="s2"] > summary')
+  await p.click('#activity .subagent[data-id="s2"] .subagent__buttons .btn >> text=Open its conversation')
+  await p.waitForSelector("#activity .convo .convo__turn")
+  const convo = await p.textContent("#activity-body")
+  assert.match(convo, /Write the routes\./)
+  assert.match(convo, /Reading the API first\./)
+  assert.match(convo, /src\/api\.py/)
+  assert.equal(await p.locator("#activity-body textarea, #activity-body input").count(), 0, "nothing to type into")
+  await p.click("#activity-back")
+  await p.click('#activity .subagent[data-id="s2"] .subagent__buttons .btn >> text=Findings')
+  await p.waitForSelector("#activity .findings")
+  assert.match(await p.textContent("#activity .findings"), /Login is POST \/session\./)
+  await p.click("#activity-back")
+  await p.click('#activity .subagent[data-id="s2"] button.agent__stop')
+  await p.waitForSelector("dialog[open]")
+  assert.match(await p.textContent("dialog[open]"), /Stop routes and the subagents under it\?/)
+  await p.click("dialog[open] .btn--danger")
+  await p.waitForFunction(() => window.__mock.api("POST", "/chats/chat2/subagents/s2/stop").length === 1)
+  assert.deepEqual(p.errors, [])
+})
+
+test("C4: the stream's tree frame redraws the panel; with the device away the kept tree shows and the rest says why", async () => {
+  const away = { ...fixture("tree"), live: false }
+  const p = await page(c4({ "GET /chats/chat2/tree": away, "GET /chats/chat2/subagents/s2/transcript": { __error: { status: 409, detail: "That is kept on the computer this chat runs on, which is not linked right now." } } }))
+  await openTree(p)
+  assert.match(await p.textContent("#activity .subagents"), /This device is not linked right now\./)
+  await p.click('#activity .subagent[data-id="s2"] > summary')
+  await p.click('#activity .subagent[data-id="s2"] .subagent__buttons .btn >> text=Open its conversation')
+  await p.waitForSelector("#activity-body .is-error")
+  assert.match(await p.textContent("#activity-body"), /not linked right now/)
+  await p.click("#activity-back")
+  // A reply streams; its `tree` frame (event_tree.json) is the tree now.
+  await p.evaluate(() => window.__mock.frame("/chats/chat2/messages/m2/stream", "tree", null))
+  await p.context().close()
+  const live = await page({ ...c4(), routes: { ...c4().routes, "GET /chats/chat2": { chat: agentChat(), messages: [userMessage("m1", "Go")], more: false, live: "m2" } }, streams: { [STREAM2]: [] } })
+  await openTree(live)
+  await live.evaluate(([path, f]) => window.__mock.frame(path, "tree", f), [STREAM2, fixture("event_tree")])
+  await live.waitForFunction(() => !!document.querySelector('#activity .subagent[data-id="s2"]') && /parser/.test(document.querySelector('#activity .subagent[data-id="s2"]').textContent))
+  assert.match(await live.textContent('#activity .subagent[data-id="s2"]'), /reviews the parser/)
+  assert.match(await live.textContent('#activity .subagent[data-id="s2"]'), /manual \+ write/)
+  assert.deepEqual(live.errors, [])
+})
+
+test("C4: an approval a subagent asks for names it", async () => {
+  const card = fixture("event_ask").cards.find((c) => c.id === "card4")
+  const p = await page(agentChatOpen([["ask", { cards: [card] }]], { "POST /chats/chat2/interactions/card4": { answered: true } }))
+  await openChatRow(p, "chat2")
+  await p.waitForSelector("#cards .ask__subagent")
+  assert.equal(await p.textContent("#cards .ask__subagent"), "Subagent parser (s1), reviews the parser, asks:")
+  assert.equal(await p.inputValue("#cards .ask__edit"), "git push", "the command, correctable, as for the main agent's")
+  assert.deepEqual(p.errors, [])
+})
+
+test("C4: Subagents for a new agent chat as /models?kind=agent says for this device; for an open one as the chat says", async () => {
+  const facts = fixture("models_agent")
+  const newAgent = async (models) => {
+    const p = await page(c4({ "GET /models?kind=agent&device=token1": models }))
+    await p.click("#new-chat-btn")
+    await p.click(".kind__tab >> text=Agent")
+    await p.waitForSelector("#composer-chips .cchip")
+    await p.waitForTimeout(200)
+    return p
+  }
+  let p = await newAgent(facts)
+  assert.equal(await p.locator("#composer-chips .cchip--subagents").count(), 1, "offered, and the device runs them")
+  await p.click("#composer-chips .cchip--subagents")
+  await p.waitForSelector(".range2")
+  assert.equal(await p.getAttribute(".range2__input--min", "max"), "32", "the server's cap")
+  await p.context().close()
+  p = await newAgent({ ...facts, agent_subagents: { offered: true, device_runs: false } })
+  assert.equal(await p.locator("#composer-chips .cchip--subagents").count(), 0, "this computer's CLI cannot run them")
+  await p.context().close()
+  p = await newAgent({ ...facts, agent_subagents: { offered: false, device_runs: true } })
+  assert.equal(await p.locator("#composer-chips .cchip--subagents").count(), 0, "not offered")
+  await p.context().close()
+  p = await newAgent({ __error: { status: 400, detail: "older server" } })
+  assert.equal(await p.locator("#composer-chips .cchip--subagents").count(), 0, "a server before 3.4.0: none, as before")
+  await p.context().close()
+  // An open agent chat: its own `subagents` (chat_agent.json offers them).
+  const chat = fixture("chat_agent")
+  p = await page(c4({ "GET /chats/chat2": { chat: { ...agentChat({ running: false }), subagents: chat.subagents, agents: chat.agents }, messages: [userMessage("m1", "Go")], more: false, live: null } }))
+  await openChatRow(p, "chat2")
+  await p.waitForSelector("#composer-chips .cchip--subagents")
+  assert.deepEqual(p.errors, [])
+})
+
+test("C4: the tree at 390 px stays inside the window", async () => {
+  const p = await page(c4(), { width: 390, height: 844 })
+  await p.evaluate(() => document.querySelector('.chatrow[data-id="chat2"]').click())
+  await p.waitForSelector("#activity-open")
+  await p.click("#activity-open")
+  await p.waitForSelector("#activity .subagents .subagent")
+  await p.click('#activity .subagent[data-id="s2"] > summary')
+  assert.ok(await p.evaluate(() => document.documentElement.scrollWidth <= 390), "nothing scrolls sideways")
+  const panel = await p.evaluate(() => { const b = document.querySelector("#activity-body"); return b.scrollWidth <= b.clientWidth })
+  assert.ok(panel, "the panel does not scroll sideways")
+  if (process.env.SHOTS) await p.screenshot({ path: join(process.env.SHOTS, "c4-tree-390.png") })
+  assert.deepEqual(p.errors, [])
+})
+
+// --- Run ----------------------------------------------------------------------------------------
+
+const only = process.argv[2] || ""
+browser = await launch()
+let failed = 0
+for (const { name, fn } of tests) {
+  if (only && !name.includes(only)) continue
+  try {
+    await fn()
+    console.log(`ok   ${name}`)
+  } catch (err) {
+    failed++
+    console.log(`FAIL ${name}\n     ${String(err?.stack || err).split("\n").slice(0, 6).join("\n     ")}`)
+  }
+}
+await browser.close()
+console.log(failed ? `tests/ui: ${failed} failed` : "tests/ui: all passed")
+process.exit(failed ? 1 : 0)

tokens.lock.json

@@ -1,5 +1,5 @@
 {
-  "tag": "v2.0.0-beta.6",
+  "tag": "v2.0.0-beta.8",
-  "commit": "a71e9b4bdb8acb6e9949e978253a72c347cc67f3",
+  "commit": "d06ca353b801916302f0ce0cc9ae3e5244ff184f",
-  "sha256": "e5f76f542ed5512702e06d69fa1dfa098c511f67dacbd7cc93f4a0618c6ae966"
+  "sha256": "6fd6d9d530c005dc52ad7dfcd61b3515b29955aa6d55365d35df3936fb7f29da"
 }

ui/app.css

@@ -122,6 +122,7 @@ textarea { padding: var(--sp-2) var(--control-px-sm); resize: none; line-height:
 
 label, .label { font: var(--text-sm) / var(--leading-tight) var(--font-body); color: var(--ink-muted); }
 .hint { font-size: var(--text-sm); color: var(--ink-muted); min-height: calc(var(--text-sm) * var(--leading-normal)); }
+.hint.is-warn { color: var(--warning); }
 .status { font: var(--text-sm) / var(--leading-normal) var(--font-body); }
 .status.is-ok { color: var(--leaf); }
 .check { display: grid; grid-template-columns: var(--check-size) minmax(0, 1fr); gap: var(--sp-2); align-items: center; color: var(--ink); }
@@ -326,6 +327,8 @@ label, .label { font: var(--text-sm) / var(--leading-tight) var(--font-body); co
 .shell[data-side="false"] { --side-col: 0px; }
 .shell[data-viewer="true"] { --panel-col: minmax(var(--inspector-width-min), 42%); }
 .shell[data-activity="true"] { --panel-col: var(--inspector-width); }
+/* The terminal: the server's width for it, giving way down to its least before the chat does. */
+.shell[data-terminal="true"] { --panel-col: minmax(var(--terminal-width-min), var(--terminal-width)); }
 
 /* --- The sidebar --------------------------------------------------------------------------- */
 
@@ -708,26 +711,75 @@ details[open] > summary .step__chev .icon { transform: rotate(90deg); }
 }
 .toggle input:checked { background: var(--accent); border-color: var(--accent); }
 .toggle input:checked::after { background: var(--accent-ink); transform: translateX(var(--check-size)); }
-/* The team's size: two number fields of one width with a dash between them, and in the header
+/* The subagents, Min and Max: one range with two handles. Three rows of known height -- the
-   its word before them -- fixed tracks, so nothing moves when a number grows a digit. */
+   values said on one line, the track box (one control high), the two ends -- and in the box two
-.agents {
+   range inputs laid over one track, the span between the handles filled. The inputs let only
-  display: inline-grid;
+   their handles take the pointer, so either can be grabbed wherever the other is. */
-  grid-template-columns: minmax(0, 1fr) auto minmax(0, 1fr);
+.range2 { display: grid; grid-template-rows: auto var(--control-h) auto; gap: var(--sp-1); width: 100%; }
-  gap: var(--sp-2);
+.range2__said { font: 600 var(--text-sm) / var(--leading-tight) var(--font-ui); color: var(--ink); font-variant-numeric: tabular-nums; }
-  align-items: center;
+.range2__box { position: relative; display: block; height: var(--control-h); --lo: 0%; --hi: 100%; }
+.range2__track {
+  position: absolute;
+  top: 50%;
+  left: calc(var(--check-size) / 2);
+  right: calc(var(--check-size) / 2);
+  height: var(--sp-1);
+  transform: translateY(-50%);
+  background: var(--border);
+  border-radius: var(--radius-full);
+}
+.range2__fill {
+  position: absolute;
+  top: 0;
+  bottom: 0;
+  left: var(--lo);
+  right: calc(100% - var(--hi));
+  background: var(--accent);
+  border-radius: var(--radius-full);
+}
+.range2__input {
+  position: absolute;
+  inset: 0;
   width: 100%;
-  height: var(--control-h);
+  height: 100%;
+  margin: 0;
+  background: none;
+  pointer-events: none;
+  -webkit-appearance: none;
+  appearance: none;
 }
-.agents__dash { color: var(--ink-muted); }
+.range2__input.is-top { z-index: 1; }
-.agents__input { width: 100%; min-width: 0; height: var(--control-h); padding: 0 var(--control-px-sm); }
+.range2__input::-webkit-slider-runnable-track { height: 100%; background: none; border: 0; }
-.agents--compact {
+.range2__input::-moz-range-track { height: 100%; background: none; border: 0; }
-  grid-template-columns: auto calc(var(--w-field-xs) / 2) auto calc(var(--w-field-xs) / 2);
+.range2__input::-webkit-slider-thumb {
-  width: auto;
+  -webkit-appearance: none;
-  height: var(--control-h-sm);
+  appearance: none;
-  font: var(--text-sm) var(--font-ui);
+  width: var(--check-size);
+  height: var(--check-size);
+  margin-top: calc((var(--control-h) - var(--check-size)) / 2);
+  background: var(--surface-raised);
+  border: var(--border-w-thick) solid var(--accent);
+  border-radius: var(--radius-full);
+  box-shadow: var(--shadow-sm);
+  cursor: grab;
+  pointer-events: auto;
 }
-.agents--compact .agents__input { height: var(--control-h-sm); }
+.range2__input::-moz-range-thumb {
-.agents__label { color: var(--ink-muted); white-space: nowrap; }
+  width: var(--check-size);
+  height: var(--check-size);
+  background: var(--surface-raised);
+  border: var(--border-w-thick) solid var(--accent);
+  border-radius: var(--radius-full);
+  box-shadow: var(--shadow-sm);
+  cursor: grab;
+  pointer-events: auto;
+}
+.range2__input:focus-visible { outline: none; }
+.range2__input:focus-visible::-webkit-slider-thumb { box-shadow: var(--ring); }
+.range2__input:focus-visible::-moz-range-thumb { box-shadow: var(--ring); }
+/* The two ends, each one handle wide, so each number sits centred under its handle at that end. */
+.range2__ends { display: flex; justify-content: space-between; font-size: var(--text-xs); color: var(--ink-muted); font-variant-numeric: tabular-nums; }
+.range2__ends > span { width: var(--check-size); text-align: center; white-space: nowrap; }
 /* Markdown, drawn by markdown.js from model output (never as HTML). */
 .md { overflow-wrap: anywhere; min-width: 0; }
 .md > * + * { margin-top: var(--sp-3); }
@@ -920,6 +972,14 @@ details[open] > summary .step__chev .icon { transform: rotate(90deg); }
 .ctx__name { overflow: hidden; white-space: nowrap; text-overflow: ellipsis; color: var(--ink-muted); }
 .ctx__num { font: var(--text-xs) var(--font-mono); text-align: right; color: var(--ink-muted); }
 .ctx__note { margin: 0; font-size: var(--text-xs); color: var(--ink-faint); }
+/* "41k until auto-compact" and the Compact button: the words give, the button keeps its size. */
+.ctx__compact { display: grid; grid-template-columns: minmax(0, 1fr) auto; gap: var(--sp-1) var(--sp-2); align-items: center; padding-top: var(--sp-2); border-top: var(--border-w) solid var(--border); }
+.ctx__until { color: var(--ink-muted); }
+.ctx__error { grid-column: 1 / -1; margin: 0; font-size: var(--text-xs); color: var(--warning); }
+/* Environment: an agent chat's folder and where it is. */
+.env { display: grid; grid-template-columns: var(--sp-4) minmax(0, auto) auto; justify-content: start; gap: var(--sp-2); align-items: center; margin: 0; }
+.env__name { overflow: hidden; white-space: nowrap; text-overflow: ellipsis; color: var(--ink); }
+.env__where { font: var(--text-xs) var(--font-mono); color: var(--ink-muted); }
 .ctx--messages { --ctx: var(--accent); }
 .ctx--system { --ctx: var(--ink-muted); }
 .ctx--memory { --ctx: var(--warning); }
@@ -943,15 +1003,41 @@ details[open] > summary .step__chev .icon { transform: rotate(90deg); }
    track that gives, cut), elapsed time in a column of its own width, the stop button at its size
    (an empty box where there is none) -- so rows line up whatever each says. */
 .tasks { display: grid; gap: var(--sp-2); min-width: 0; }
-.tasks__summary, .tasks__team { margin: 0; color: var(--ink-muted); }
+.tasks__team { margin: 0; color: var(--ink-muted); }
 .tasks__goal { margin: 0; color: var(--ink); }
-.tasks__list { display: grid; gap: var(--sp-1); }
+.tasks__group { display: grid; grid-template-columns: auto auto; justify-content: start; gap: var(--sp-2); align-items: baseline; margin: var(--sp-1) 0 0; color: var(--ink-muted); }
-.tasks__more > summary { cursor: pointer; padding: var(--sp-1) 0; color: var(--ink-muted); }
+.tasks__list { display: grid; gap: var(--sp-2); }
-.tasks__more[open] > summary { margin-bottom: var(--sp-1); }
+/* A fold ("Show 3 more", "Finished 12", a task's own "2 tasks"): a chevron that turns, then the
+   words; nested ones step in by the depth of what is under them. */
+.tasks__more { margin-left: calc(var(--depth, 0) * var(--sp-4)); min-width: 0; }
+.tasks__more > summary {
+  display: grid;
+  grid-template-columns: 12px minmax(0, 1fr);
+  gap: var(--sp-2);
+  align-items: center;
+  min-height: var(--row-h);
+  padding: 0 var(--sp-3);
+  color: var(--ink-muted);
+  cursor: pointer;
+  list-style: none;
+}
+.tasks__more > summary::-webkit-details-marker { display: none; }
+.tasks__more > summary:hover { color: var(--ink); }
+.tasks__more[open] > summary { margin-bottom: var(--sp-2); }
+.tasks__nested { margin-top: calc(-1 * var(--sp-1)); }
+.tasks__finished { padding-top: var(--sp-1); border-top: var(--border-w) solid var(--border); }
+.tasks__chev, .agent__chev { display: grid; place-items: center; width: 12px; height: 12px; color: var(--ink-faint); transition: transform var(--dur-1) var(--ease-out); }
+.tasks__chev .icon, .agent__chev .icon { width: 12px; height: 12px; }
+.tasks__more[open] > summary .tasks__chev, .agent[open] > .agent__row .agent__chev { transform: rotate(90deg); }
+@media (prefers-reduced-motion: reduce) { .tasks__chev, .agent__chev { transition: none; } }
+
+/* A task: one card, one row of known tracks -- the chevron, what it does (the one track that
+   gives, cut), the elapsed time in a column of its own width, the stop square at its size (an
+   empty box where it cannot be stopped) -- so the rows line up whatever each says. */
 .agent { margin-left: calc(var(--depth, 0) * var(--sp-4)); background: var(--surface); border: var(--border-w) solid var(--border); border-radius: var(--radius-lg); min-width: 0; }
 .agent__row {
   display: grid;
-  grid-template-columns: var(--dot-size) minmax(0, 1fr) var(--time-col) var(--control-h-sm);
+  grid-template-columns: 12px minmax(0, 1fr) var(--time-col) var(--control-h-sm);
   gap: var(--sp-2);
   align-items: center;
   min-height: var(--control-h);
@@ -960,17 +1046,21 @@ details[open] > summary .step__chev .icon { transform: rotate(90deg); }
   list-style: none;
 }
 .agent__row::-webkit-details-marker { display: none; }
-.agent__dot { width: var(--dot-size); height: var(--dot-size); border-radius: var(--radius-full); background: var(--ink-faint); }
+.agent__row:hover { background: var(--surface-hover); border-radius: var(--radius-lg); }
-.agent.is-working .agent__dot { background: var(--accent); }
-.agent.is-queued .agent__dot, .agent.is-waiting .agent__dot { background: var(--warning); }
-.agent.is-done .agent__dot { background: var(--leaf); }
-.agent.is-error .agent__dot { background: var(--danger); }
 .agent__what { display: grid; min-width: 0; }
-.agent__name, .agent__role { overflow: hidden; white-space: nowrap; text-overflow: ellipsis; }
+.agent__name { overflow: hidden; white-space: nowrap; text-overflow: ellipsis; color: var(--ink); }
-.agent__name { color: var(--ink); }
+.agent.is-done .agent__name, .agent.is-stopped .agent__name { color: var(--ink-muted); }
-.agent__role { font-size: var(--text-xs); color: var(--ink-muted); }
+.agent.is-error .agent__name { color: var(--danger); }
+.agent__role { margin: 0; color: var(--ink-muted); overflow-wrap: anywhere; }
 .agent__time { font: var(--text-xs) var(--font-mono); text-align: right; color: var(--ink-muted); }
-.agent__stop { width: var(--control-h-sm); height: var(--control-h-sm); }
+.agent.is-queued .agent__time, .agent.is-waiting .agent__time { color: var(--warning); }
+/* The stop square: a small filled square in a button the size of the others. */
+.agent__stop { display: grid; place-items: center; width: var(--control-h-sm); height: var(--control-h-sm); padding: 0; background: var(--surface-active); border: 0; border-radius: var(--radius-md); cursor: pointer; }
+.agent__stop:hover { background: var(--surface-hover); }
+.agent__stop.is-none { background: none; cursor: default; }
+.agent__square { width: var(--sp-3); height: var(--sp-3); background: var(--ink); border-radius: var(--radius-sm); }
+.agent__stop:hover .agent__square { background: var(--danger); }
+.agent__id { font: var(--text-xs) var(--font-mono); color: var(--ink-faint); overflow-wrap: anywhere; }
 .agent__body { display: grid; gap: var(--sp-2); padding: var(--sp-2) var(--sp-3) var(--sp-3); border-top: var(--border-w) solid var(--border); min-width: 0; }
 .agent__body > .btn { justify-self: start; }
 .agent__facts { display: flex; flex-wrap: wrap; gap: var(--sp-1) var(--sp-3); margin: 0; font-size: var(--text-xs); color: var(--ink-muted); }
@@ -1203,6 +1293,7 @@ details[open] > summary .step__chev .icon { transform: rotate(90deg); }
 }
 .cchip:hover { color: var(--ink); background: var(--surface-hover); border-color: var(--border); }
 .cchip:disabled { cursor: default; }
+.cchip[aria-pressed="true"], .tool[aria-pressed="true"] { color: var(--ink); background: var(--surface-active); }
 .cchip:disabled:hover { background: transparent; border-color: transparent; }
 .cchip__label { overflow: hidden; text-overflow: ellipsis; }
 .cchip__chev { display: grid; color: var(--ink-faint); }
@@ -1470,12 +1561,80 @@ details[open] > summary .step__chev .icon { transform: rotate(90deg); }
 .folders .path { overflow: hidden; white-space: nowrap; text-overflow: ellipsis; direction: rtl; text-align: left; font: var(--text-sm) var(--font-ui); }
 .folders .empty { display: block; padding: var(--sp-3); color: var(--warning); font-size: var(--text-sm); text-align: left; }
 
+/* --- The terminal panel -------------------------------------------------------------------------
+   The panel column, as the activity panel has it, and never beside it (opening one closes the
+   other; the shell lives on). The viewer goes over the chat while it is open, as with the
+   activity panel. Not `.terminal`: xterm gives its own element that class. The head is the
+   viewer's: an icon, the device's name over the folder, then the buttons on known tracks (the
+   command ones only while the shell marks its commands). */
+
+.shell[data-terminal="true"] .viewer { grid-area: main; z-index: var(--z-panel); border-left: 0; }
+.termpanel {
+  grid-area: panel;
+  min-width: 0;
+  min-height: 0;
+  display: grid;
+  grid-template-rows: auto minmax(0, 1fr) auto;
+  border-left: var(--border-w) solid var(--border);
+  background: var(--bg-sunken);
+}
+.termpanel__head {
+  grid-template-columns: 16px minmax(0, 1fr);
+  grid-auto-flow: column;
+  grid-auto-columns: var(--control-h-sm);
+}
+.termpanel__icon { display: grid; place-items: center; color: var(--ink-muted); }
+.termpanel__where { font-family: var(--font-mono); }
+/* xterm measures this box: it has a size of its own, and the terminal scrolls inside it. */
+.termpanel__screen { min-height: 0; min-width: 0; overflow: hidden; padding: var(--sp-2); background: var(--code-bg); }
+.termpanel__screen .xterm { height: 100%; }
+/* What happened, the last command, and Send: the words take what is left and wrap; the last
+   command is one clipped line; the button keeps its size. */
+.termpanel__foot {
+  display: grid;
+  grid-template-columns: minmax(0, 1fr);
+  grid-auto-flow: column;
+  grid-auto-columns: auto;
+  gap: var(--sp-2);
+  align-items: center;
+  padding: var(--sp-2) var(--sp-3);
+  border-top: var(--border-w) solid var(--border);
+  font-size: var(--text-xs);
+  color: var(--ink-faint);
+}
+.termpanel__message { margin: 0; min-width: 0; overflow-wrap: anywhere; }
+.termpanel__message.is-error { color: var(--danger); }
+.termpanel__last {
+  max-width: var(--w-chip);
+  font-family: var(--font-mono);
+  color: var(--ink-muted);
+  white-space: nowrap;
+  overflow: hidden;
+  text-overflow: ellipsis;
+}
+.termpanel__last:empty { display: none; }
+
+/* --- The subagent tree (spec 3.4.0) ------------------------------------------------------------
+   Rows are the tasks' rows (.agent): name and id on one line, the position, then the phase (or how
+   it ended) clipped to one line -- its clock is in the opened row. Ended ones are dimmed. */
+
+.subagents { display: grid; gap: var(--sp-2); min-width: 0; }
+.subagents .tasks__team, .subagents .tasks__goal { margin: 0; }
+.subagent__who { display: flex; gap: var(--sp-2); align-items: baseline; min-width: 0; }
+.subagent__who .agent__id { flex: none; }
+.subagent__phase { overflow: hidden; white-space: nowrap; text-overflow: ellipsis; font-size: var(--text-xs); color: var(--ink-muted); }
+.agent.is-ended { opacity: var(--opacity-dim); }
+.subagent__buttons { display: flex; flex-wrap: wrap; gap: var(--sp-2); margin: 0; }
+.findings { margin: 0; padding: var(--sp-3) var(--sp-3) var(--sp-3) calc(var(--sp-3) + var(--sp-5)); display: grid; gap: var(--sp-3); }
+.findings__item { min-width: 0; overflow-wrap: anywhere; }
+.ask__subagent { color: var(--ink); }
+
 /* --- Narrower windows ---------------------------------------------------------------------- */
 
 /* 64rem (--bp-wide): a panel goes over the chat instead of beside it. */
 @media (max-width: 64rem) {
-  .shell[data-viewer="true"], .shell[data-activity="true"] { --panel-col: 0px; }
+  .shell[data-viewer="true"], .shell[data-activity="true"], .shell[data-terminal="true"] { --panel-col: 0px; }
-  .viewer, .activity, .shell[data-activity="true"] .viewer { grid-area: main; z-index: var(--z-panel); border-left: 0; }
+  .viewer, .activity, .termpanel, .shell[data-activity="true"] .viewer, .shell[data-terminal="true"] .viewer { grid-area: main; z-index: var(--z-panel); border-left: 0; }
   .task__open { grid-template-columns: var(--sp-8) minmax(0, 1fr) var(--w-field-sm); }
   .task__next { display: none; }
 }

ui/index.html

@@ -135,6 +135,26 @@
       </div>
     </section>
 
+    <aside class="termpanel" id="terminal" hidden aria-labelledby="terminal-title">
+      <header class="viewer__head termpanel__head">
+        <span class="termpanel__icon" id="terminal-icon" aria-hidden="true"></span>
+        <div class="viewer__id">
+          <h2 class="viewer__name" id="terminal-title" data-i18n>Terminal</h2>
+          <p class="viewer__where termpanel__where" id="terminal-where"></p>
+        </div>
+        <button type="button" class="tool" id="terminal-prev" hidden aria-label="The previous command" title="The previous command (Ctrl+Shift+↑)" data-i18n-aria-label data-i18n-title></button>
+        <button type="button" class="tool" id="terminal-next" hidden aria-label="The next command" title="The next command (Ctrl+Shift+↓)" data-i18n-aria-label data-i18n-title></button>
+        <button type="button" class="tool" id="terminal-copy" aria-label="Copy the last command and its output" title="Copy the last command and its output" data-i18n-aria-label data-i18n-title></button>
+        <button type="button" class="tool" id="terminal-close" aria-label="Close terminal" title="Close terminal" data-i18n-aria-label data-i18n-title></button>
+      </header>
+      <div class="termpanel__screen" id="terminal-screen"></div>
+      <footer class="termpanel__foot">
+        <p class="termpanel__message" id="terminal-message" aria-live="polite"></p>
+        <span class="termpanel__last" id="terminal-last"></span>
+        <button type="button" class="btn btn--sm termpanel__send" id="terminal-send" title="Put the last command and its output into the message box" data-i18n-title><span data-i18n>Send to chat</span></button>
+      </footer>
+    </aside>
+
     <aside class="activity" id="activity" hidden aria-labelledby="activity-title">
       <header class="viewer__head activity__head">
         <button type="button" class="tool" id="activity-back" hidden aria-label="Back" title="Back" data-i18n-aria-label data-i18n-title></button>

ui/js/activity.js

@@ -1,14 +1,26 @@
-// The chat's activity panel, beside it: what is in its context, by category, against the model's
+// The chat's activity panel, beside it, opened from the header's button at any time: where an
-// window; the files the chat and its agents wrote; every background agent -- a team's members as
+// agent chat works; what is in its context, by category, against the model's window, how far it
-// a tree, helpers and commands -- with its state, how long it has run, its model, its latest step
+// is to the automatic compaction and a button to compact now; the files the chat and its
-// and what it is writing, its whole conversation read-only and a button to stop it; the team's
+// subagents wrote; every background task -- subagents, a team's members as a tree under the one
-// room; and the plan. The strip above the message box sums the agents up and opens it. The same
+// that leads them, background commands -- with how long it has run and a stop square, unfolding
-// facts as the server's own panel (`services/activity.py`), from /api/v2:
+// to its latest step, what it is writing and its whole conversation, read-only; the finished ones
+// folded; the team's room; and the plan. The line above the message box sums up what runs, while
+// anything does, and opens it. The same facts as the server's own panel (`services/activity.py`),
+// from /api/v2:
 //
 //   GET  /chats/{chat}/activity?since=&context=   everything, with a `version`
 //   GET  /chats/{chat}/activity/events            `activity` {version, summary} on each change
-//   GET  /chats/{chat}/agents/{kind}/{id}         one agent and its conversation
+//   GET  /chats/{chat}/agents/{kind}/{id}         one task and its conversation
 //   POST /chats/{chat}/agents/{kind}/{id}/stop
+//   POST /chats/{chat}/compact                    summarise the earlier turns now
+//   GET  /chats/{chat}/tree?since=                the subagent tree (spec 3.4.0) and the room
+//   GET  /chats/{chat}/subagents/{id}/transcript  a subagent's conversation, read-only
+//   GET  /chats/{chat}/subagents/{id}/findings    what it kept
+//   POST /chats/{chat}/subagents/{id}/stop        it and those under it
+//
+// The tree comes too as the reply stream's `tree` frame (chat.js hands it on). An agent chat's
+// subagents run on its device: while the device is away the server sends the tree it kept
+// (`live` false) and answers 409 for a conversation or findings, in its own words.
 //
 // The events stream is followed for as long as a chat is open: it keeps the strip current, and
 // while the panel is open a new `version` reads the panel again. The context is the costly part
@@ -23,6 +35,7 @@ import { get, post, follow, why } from "./bridge.js"
 import { markdown } from "./markdown.js"
 import { ask } from "./popup.js"
 import { openServerFile } from "./explorer.js"
+import { treePart, tickTree, treeState, TREE_LIVE } from "./tree.js"
 
 const ACTIVE = ["queued", "working", "waiting"]
 // How many live agents show before "Show N more", as on the server's panel.
@@ -31,6 +44,8 @@ const SHOWN = 6
 const CONTEXT_EVERY = 15000
 
 let layout = () => {} // (on) -> the shell makes room for the panel
+let onThisComputer = () => false // (chat) -> whether an agent chat works on this computer
+let onCompacted = () => {} // (chat id) -> the conversation is read again
 let chat = null // the open chat: { id, kind, … }
 let stream = null // the events stream's handle
 let retry = 0 // the reconnect timer
@@ -44,7 +59,12 @@ let readAt = 0 // when `data` was read: the elapsed times count on from there
 let contextAt = 0
 let reading = null // the read in flight
 let readAgain = "" // "", "plain" or "context": what to read again once the read in flight is done
-let view = null // null: the panel; { kind, id, name }: one agent's conversation
+let view = null // null: the panel; { kind, id, name }: one agent's conversation ({ kind: "sub", what } a subagent's)
+let tree = null // the subagent tree (spec 3.4.0), null for none or an older server
+let treeRoom = [] // the room as the tree's answer gives it (an agent chat's comes only there)
+let treeLive = true // false: an agent chat's device is away, and this is the tree the server kept
+let treeNote = "" // why the last stop did not go through
+let subTimer = 0 // a live subagent's conversation, read again while it works
 const openRows = new Set() // agents unfolded, by kind:id; sections folded, by name
 let ticker = 0
 
@@ -69,12 +89,14 @@ export function compact(value) {
   return `${trim(n / 1000000, 2)}M`
 }
 
-/** "3 agents working · 1 queued · 2 finished", or "" for none (the server's `summary_text`). */
+/**
+ * "3 running · 1 queued · 2 finished", or "" for none: every background task -- a subagent, a
+ * team's member, a command -- counts the same, as the panel lists them together.
+ */
 export function summaryText(counts) {
   if (!counts) return ""
   const parts = []
-  if (counts.working === 1) parts.push(t("1 agent working"))
+  if (counts.working) parts.push(t("%(n)s running", { n: counts.working }))
-  else if (counts.working) parts.push(t("%(n)s agents working", { n: counts.working }))
   if (counts.queued) parts.push(t("%(n)s queued", { n: counts.queued }))
   if (counts.waiting) parts.push(t("%(n)s waiting", { n: counts.waiting }))
   if (counts.finished) parts.push(t("%(n)s finished", { n: counts.finished }))
@@ -139,11 +161,16 @@ export function watchActivity(next) {
   data = null
   room = []
   cursor = 0
+  tree = null
+  treeRoom = []
+  treeLive = true
+  treeNote = ""
   readAt = contextAt = 0
   view = null
+  compactError = ""
   openRows.clear()
   renderStrip()
-  $("activity-open")?.setAttribute("aria-expanded", String(panelOpen()))
+  markOpen()
   if (!chat) {
     if (panelOpen()) showPanel(false)
     return
@@ -155,6 +182,15 @@ export function watchActivity(next) {
   }
 }
 
+/** The reply stream's `tree` frame: the tree as it is now. */
+export function treeChanged(chatId, next) {
+  if (!chat || chat.id !== chatId) return
+  tree = next || null
+  if (!panelOpen()) return
+  if (!view) render()
+  else if (view.kind === "sub" && view.what === "transcript") readAgent()
+}
+
 /** A reply has ended, or the chat's settings changed: the context and the outputs move with it. */
 export function activityChanged() {
   if (chat && panelOpen()) read(true)
@@ -232,8 +268,16 @@ async function read(withContext) {
   const id = chat.id
   reading = (async () => {
     try {
-      const a = (await get(`/chats/${enc(id)}/activity?since=${cursor}&context=${withContext ? "true" : "false"}`)).activity
+      const [r, tr] = await Promise.all([
+        get(`/chats/${enc(id)}/activity?since=${cursor}&context=${withContext ? "true" : "false"}`),
+        // A server before 3.4.0 has no tree: 404, and the panel is as it was.
+        get(`/chats/${enc(id)}/tree`).catch(() => null),
+      ])
+      const a = r.activity
       if (chat?.id !== id || !a) return
+      tree = tr?.tree || null
+      treeRoom = tr?.room || []
+      treeLive = tr ? tr.live !== false : true
       if (!withContext) a.context = data?.context ?? null
       else contextAt = Date.now()
       if (a.team) {
@@ -269,36 +313,58 @@ async function read(withContext) {
   }
 }
 
-// --- The strip above the message box ---------------------------------------------------------------
+// --- The line above the message box, and the header's button ---------------------------------------
+
+/** Both ways into the panel say whether it is open. */
+function markOpen() {
+  for (const id of ["activity-open", "activity-strip-btn"]) $(id)?.setAttribute("aria-expanded", String(panelOpen()))
+}
 
 function renderStrip() {
   const strip = $("activity-strip")
-  // A chip for every open chat: what runs, while something does, as the server's strip says it (a
+  // Only while something runs, as the server's own line: the header's button is the way into the
-  // chat whose helpers all finished long ago does not carry "12 finished" for ever: the panel
+  // panel at any other time (and a chat whose tasks all finished long ago does not carry
-  // still lists them); else only the way into the panel.
+  // "12 finished" for ever: the panel still lists them).
-  strip.hidden = !chat
+  const active = chat && summary?.active ? summaryText(summary) : ""
-  if (!chat) return clear(strip)
+  strip.hidden = !active
-  const active = summary?.active ? summaryText(summary) : ""
+  if (!active) return clear(strip)
-  const text = active || t("Activity")
   strip.replaceChildren(
     h(
       "button",
       {
         type: "button",
-        id: "activity-open",
+        id: "activity-strip-btn",
-        class: `strip__btn${active ? " is-active" : ""}`,
+        class: "strip__btn is-active",
-        "aria-label": `${text} — ${panelOpen() ? t("Close the activity panel") : t("Open the activity panel")}`,
+        "aria-label": `${active} — ${panelOpen() ? t("Close the activity panel") : t("Open the activity panel")}`,
         "aria-controls": "activity",
         "aria-expanded": String(panelOpen()),
         onclick: togglePanel,
       },
       h("span", { class: "dot", "aria-hidden": "true" }),
-      h("span", { class: "strip__text", text }),
+      h("span", { class: "strip__text", text: active }),
       icon(ICONS.chevron),
     ),
   )
 }
 
+/** The header's button: the panel, open or closed, for every open chat. */
+export function activityButton() {
+  return h(
+    "button",
+    {
+      type: "button",
+      class: "tool",
+      id: "activity-open",
+      "aria-label": t("Context and background tasks"),
+      title: t("Context and background tasks"),
+      "aria-controls": "activity",
+      "aria-expanded": String(panelOpen()),
+      onclick: togglePanel,
+    },
+    icon(ICONS.tasks),
+  )
+}
+
 // --- The panel ------------------------------------------------------------------------------------------
 
 /** Open or close the panel (the header's Activity button, the strip, Escape). */
@@ -306,14 +372,16 @@ export function showPanel(on) {
   if (on && !chat) return
   $("activity").hidden = !on
   layout(on)
-  $("activity-open")?.setAttribute("aria-expanded", String(on))
+  markOpen()
   clearInterval(ticker)
   ticker = 0
   if (!on) {
     view = null
+    renderStrip()
     return
   }
   ticker = setInterval(tick, 1000)
+  renderStrip()
   render()
   read(true)
   $("activity-close").focus()
@@ -330,6 +398,7 @@ function tick() {
   for (const el of document.querySelectorAll("#activity .agent__time[data-active]")) {
     el.textContent = duration(Number(el.dataset.seconds) + extra)
   }
+  tickTree($("activity-body"))
 }
 
 function renderError(text) {
@@ -359,15 +428,45 @@ function render() {
   }
   // Kept where the reader was: the panel is drawn again on every change.
   const scroll = body.scrollTop
-  body.replaceChildren(
+  // The tree draws the subagents; a team's members are not drawn again among the tasks.
+  const entries = (data.agents || []).filter((e) => !(tree && e.kind === "member"))
+  const parts = [
+    chat?.kind === "agent" && chat.device ? section("environment", t("Environment"), 0, environmentPart(chat)) : null,
     section("context", t("Context"), 0, contextPart(data.context)),
     section("outputs", t("Outputs"), (data.outputs || []).length, outputsPart(data.outputs || [])),
-    section("tasks", t("Background tasks"), 0, tasksPart()),
+    tree ? section("subagents", t("Subagents"), 0, subagentsPart()) : null,
+    section("tasks", t("Background tasks"), 0, tasksPart(entries)),
+    data.team || treeRoom.length ? section("room", t("Room"), roomList().length, roomPart()) : null,
     data.plan ? section("plan", t("Plan"), 0, h("div", { class: "activity__plan" }, planList(data.plan))) : null,
-  )
+  ]
+  body.replaceChildren(...parts.filter(Boolean))
   body.scrollTop = scroll
 }
 
+/** How many tasks, nested ones included: those still at work, or those finished. */
+function count(list, active) {
+  let n = 0
+  for (const a of list) {
+    if (ACTIVE.includes(a.state) === active) n += 1
+    n += count(a.children || [], active)
+  }
+  return n
+}
+
+/** Where an agent chat works: its folder, on this computer or another device. */
+function environmentPart(c) {
+  const dir = c.device?.directory || ""
+  const name = dir.split(/[\\/]/).filter(Boolean).pop() || dir
+  const here = onThisComputer(c)
+  return h(
+    "p",
+    { class: "env", title: dir },
+    icon(ICONS.folder),
+    h("bdi", { class: "env__name", text: name }),
+    h("span", { class: "env__where", text: here ? t("This computer") : t("Another device") }),
+  )
+}
+
 function contextPart(c) {
   if (!c) return h("p", { class: "activity__empty", text: t("Not measured.") })
   const known = c.percent !== null && c.percent !== undefined
@@ -410,7 +509,67 @@ function contextPart(c) {
     { class: "ctx__note" },
     `${c.measured ? t("Counted by the model at the last reply; the categories are estimated.") : t("Estimated: no reply has been counted yet.")} ${t("%(n)s tools offered.", { n: c.tools_count ?? 0 })}`,
   )
-  return h("div", { class: "ctx" }, used, bar, rows, note)
+  return h("div", { class: "ctx" }, used, bar, rows, compactRow(c), note)
+}
+
+/**
+ * How far it is to the automatic compaction, and the button that compacts now. The server says it
+ * (2.0.0-beta.7: `auto_compact`, `until_compact` in tokens, `compactable` whether Compact can run
+ * now); an older one gives only `threshold` -- the server compacts before a turn once the context
+ * would reach that % of the window, 0 is off -- and it is worked out from that.
+ */
+function compactRow(c) {
+  const limit = Number(c.limit) || 0
+  const threshold = Number(c.threshold)
+  const auto = "auto_compact" in c ? !!c.auto_compact : threshold > 0
+  let until = c.until_compact
+  if (until === undefined && auto && limit && threshold < 100) until = Math.floor((limit * threshold) / 100) - (Number(c.used) || 0)
+  let words
+  if (!auto) words = t("Automatic compaction is off.")
+  else if (until === null || until === undefined) words = t("Compacts by itself once the window is known.")
+  else words = until > 0 ? t("%(tokens)s until auto-compact", { tokens: compact(until) }) : t("Compacts before the next reply.")
+  const busy = compacting === chat?.id
+  // `compactable` false: nothing to summarise yet, a reply being written, compaction turned off --
+  // the button stays, greyed; pressing an older server's says why in its own words.
+  const can = !("compactable" in c) || !!c.compactable
+  return h(
+    "div",
+    { class: "ctx__compact" },
+    h("span", { class: "ctx__until", text: words }),
+    h("button", {
+      type: "button",
+      class: "btn btn--sm",
+      text: busy ? t("Compacting…") : t("Compact"),
+      disabled: busy || !can,
+      title: t("Summarise the earlier turns now and send the summary in their place."),
+      onclick: compactNow,
+    }),
+    compactError ? h("p", { class: "ctx__error", role: "alert", text: compactError }) : null,
+  )
+}
+
+let compacting = null // the chat being compacted
+let compactError = ""
+
+async function compactNow() {
+  if (!chat || compacting) return
+  const id = chat.id
+  compacting = id
+  compactError = ""
+  render()
+  try {
+    await post(`/chats/${enc(id)}/compact`)
+    compacting = null
+    if (chat?.id !== id) return
+    onCompacted(id)
+    read(true)
+  } catch (err) {
+    compacting = null
+    if (chat?.id !== id) return
+    // A server without the route in its /api/v2 says 404 or 405: said as what it means.
+    compactError = err?.status === 404 || err?.status === 405 ? t("This server cannot compact a chat from the app yet; the web interface can.") : why(err)
+    render()
+  }
 }
 
 /** A category's key as a class name: the server's words, kept to letters and underscores. */
@@ -439,64 +598,148 @@ function outputsPart(outputs) {
   )
 }
 
-function tasksPart() {
+function tasksPart(entries) {
-  const entries = data.agents || []
+  const team = tree ? null : data.team
-  const team = data.team
   const box = h("div", { class: "tasks", "aria-live": "polite" })
   if (!entries.length && !team) {
-    box.append(h("p", { class: "activity__empty", text: t("Nothing runs in the background. Helpers, commands and a team's members show here while they work.") }))
+    box.append(h("p", { class: "activity__empty", text: t("Nothing runs in the background. Subagents, a team's members and background commands show here while they work.") }))
     return box
   }
-  box.append(h("p", { class: "tasks__summary", text: summaryText(data.summary) || t("Nothing running.") }))
   if (team) {
-    box.append(h("p", { class: "tasks__team", text: t("Team: %(size)s of %(max)s agents", { size: team.size, max: team.max }) }))
+    box.append(h("p", { class: "tasks__team", text: t("Team: %(size)s of at most %(max)s subagents", { size: team.size, max: team.max }) }))
     if (team.goal) box.append(h("p", { class: "tasks__goal", text: team.goal }))
   }
-  const live = entries.filter((e) => ACTIVE.includes(e.state))
+  // At the top level, what still runs -- with its own tasks under it, finished or not -- and then
-  const ended = entries.filter((e) => !ACTIVE.includes(e.state))
+  // what finished with nothing under it still at work, folded.
+  const live = entries.filter((e) => ACTIVE.includes(e.state) || count(e.children || [], true))
+  const ended = entries.filter((e) => !live.includes(e))
+  box.append(h("p", { class: "tasks__group" }, h("span", { text: t("Running") }), h("span", { class: "activity__count", text: String(count(entries, true)) })))
   if (live.length) {
     box.append(h("div", { class: "tasks__list" }, live.slice(0, SHOWN).map(agentRows)))
     if (live.length > SHOWN) box.append(fold("more", t("Show %(n)s more", { n: live.length - SHOWN }), live.slice(SHOWN)))
-  }
+  } else box.append(h("p", { class: "activity__empty", text: t("Nothing running.") }))
-  if (ended.length) box.append(fold("finished", t("%(n)s finished", { n: ended.length }), ended))
+  if (ended.length) box.append(fold("finished", t("Finished %(n)s", { n: count(ended, false) }), ended, "tasks__finished"))
-  if (team) {
-    box.append(h("h3", { class: "activity__subheading", text: t("Room") }))
-    box.append(
-      room.length
-        ? h(
-            "ol",
-            { class: "room" },
-            room.map((m) =>
-              h(
-                "li",
-                { class: `room__entry${m.report ? " is-report" : ""}` },
-                h(
-                  "div",
-                  { class: "room__meta" },
-                  h("span", { class: "room__n", text: `#${m.n}` }),
-                  h("span", { class: "room__route", text: `${m.from} → ${(m.to || []).join(", ")}` }),
-                  m.report ? h("span", { class: "chip room__badge", text: t("report") }) : null,
-                ),
-                h("div", { class: "room__text md" }, markdown(String(m.text || ""))),
-              ),
-            ),
-          )
-        : h("p", { class: "activity__empty", text: t("Nothing has been said in the room yet.") }),
-    )
-  }
   return box
 }
 
-/** A folded list of agents ("Show 3 more", "12 finished"), remembered open across redraws. */
+/** The room: as the activity read it (a chat's team), else as the tree's answer gave it. */
-function fold(name, label, list) {
+function roomList() {
-  const el = h("details", { class: "tasks__more", open: openRows.has(`fold:${name}`) }, h("summary", { text: label }), h("div", { class: "tasks__list" }, list.map(agentRows)))
+  return room.length ? room : treeRoom
+}
+
+function subagentsPart() {
+  const box = treePart(tree, {
+    room: roomList(),
+    device: chat?.kind === "agent",
+    act: {
+      open: (a) => openSub(a, "transcript"),
+      findings: (a) => openSub(a, "findings"),
+      stop: stopSub,
+      isOpen: (key) => openRows.has(key),
+      toggled: (key, on) => (on ? openRows.add(key) : openRows.delete(key)),
+    },
+  })
+  if (!treeLive) box.prepend(h("p", { class: "activity__empty", text: t("This device is not linked right now.") }))
+  if (treeNote) box.prepend(h("p", { class: "activity__empty is-error", text: treeNote }))
+  return box
+}
+
+async function stopSub(a) {
+  if (!chat) return
+  const yes = await ask({ title: t("Stop %(name)s", { name: a.name }), text: t("Stop %(name)s and the subagents under it?", { name: a.name }), ok: t("Stop"), cancel: t("Cancel"), danger: true })
+  if (!yes) return
+  treeNote = ""
+  try {
+    await post(`/chats/${enc(chat.id)}/subagents/${enc(a.id)}/stop`)
+  } catch (err) {
+    treeNote = why(err)
+  }
+  read(false)
+}
+
+function openSub(a, what) {
+  view = { kind: "sub", id: a.id, name: a.name || a.id, what }
+  $("activity").dataset.view = "agent"
+  $("activity-title").textContent = what === "findings" ? `${view.name} · ${t("Findings")}` : view.name
+  $("activity-back").hidden = false
+  clear($("activity-body")).append(h("p", { class: "activity__empty", text: t("Loading…") }))
+  readAgent()
+}
+
+/** A subagent's conversation or findings, read-only; a live one's conversation is read again while it works. */
+async function readSub() {
+  clearTimeout(subTimer)
+  const { id, what } = view
+  let r
+  try {
+    r = await get(`/chats/${enc(chat.id)}/subagents/${enc(id)}/${what}`)
+  } catch (err) {
+    // 409: an agent chat's device is away -- the server says it is on the device.
+    if (view?.id === id && view.what === what) clear($("activity-body")).append(h("p", { class: "activity__empty is-error", text: why(err) }))
+    return
+  }
+  if (view?.kind !== "sub" || view.id !== id || view.what !== what) return
+  const body = $("activity-body")
+  if (what === "findings") {
+    const list = r.findings || []
+    body.replaceChildren(
+      list.length
+        ? h("ol", { class: "findings" }, list.map((f) => h("li", { class: "findings__item", value: String(f.n) }, h("div", { class: "md" }, markdown(String(f.text || ""))))))
+        : h("p", { class: "activity__empty", text: t("Nothing yet.") }),
+    )
+    return
+  }
+  const atEnd = body.scrollHeight - body.scrollTop - body.clientHeight < 40
+  const live = TREE_LIVE.includes(r.state)
+  const head = h("div", { class: "convo__head" }, h("p", { class: "agent__facts" }, h("span", { class: "agent__state", text: treeState(r.state) })))
+  const turns = turnsOf(r.turns || [])
+  body.replaceChildren(head, h("div", { class: "convo" }, turns.length ? turns : h("p", { class: "activity__empty", text: t("Nothing yet.") })))
+  if (atEnd) body.scrollTop = body.scrollHeight
+  if (live) subTimer = setTimeout(() => view?.kind === "sub" && view.id === id && panelOpen() && readAgent(), 3000)
+}
+
+function roomPart() {
+  const room = roomList()
+  if (!room.length) return h("p", { class: "activity__empty", text: t("Nothing has been said in the room yet.") })
+  return h(
+    "ol",
+    { class: "room" },
+    room.map((m) =>
+      h(
+        "li",
+        { class: `room__entry${m.report ? " is-report" : ""}` },
+        h(
+          "div",
+          { class: "room__meta" },
+          h("span", { class: "room__n", text: `#${m.n}` }),
+          h("span", { class: "room__route", text: `${m.from} → ${(m.to || []).join(", ")}` }),
+          m.report ? h("span", { class: "chip room__badge", text: t("report") }) : null,
+        ),
+        h("div", { class: "room__text md" }, markdown(String(m.text || ""))),
+      ),
+    ),
+  )
+}
+
+/** A folded list of tasks ("Show 3 more", "Finished 12", "Tasks: 2"), remembered across redraws. */
+function fold(name, label, list, cls = "") {
+  const el = h(
+    "details",
+    { class: `tasks__more ${cls}`, open: openRows.has(`fold:${name}`) },
+    h("summary", null, h("span", { class: "tasks__chev", "aria-hidden": "true" }, icon(ICONS.chevron)), h("span", { text: label })),
+    h("div", { class: "tasks__list" }, list.map(agentRows)),
+  )
   el.addEventListener("toggle", () => (el.open ? openRows.add(`fold:${name}`) : openRows.delete(`fold:${name}`)))
   return el
 }
 
-/** An agent and, under it, the agents it leads or started. */
+/** A task and, folded under it, the tasks it leads or started ("Tasks: 3"). */
 function agentRows(a) {
-  return [agentRow(a), ...(a.children || []).map(agentRows)]
+  const kids = a.children || []
+  if (!kids.length) return agentRow(a)
+  const nested = fold(`under:${a.kind}:${a.id}`, t("Tasks: %(n)s", { n: kids.length }), kids, "tasks__nested")
+  nested.style.setProperty("--depth", String(Math.max(0, Math.min(6, (Number(a.depth) || 0) + 1))))
+  return [agentRow(a), nested]
 }
 
 function agentRow(a) {
@@ -512,7 +755,7 @@ function agentRow(a) {
         "button",
         {
           type: "button",
-          class: "pane__btn agent__stop",
+          class: "agent__stop",
           "aria-label": t("Stop %(name)s", { name: a.name }),
           title: t("Stop %(name)s", { name: a.name }),
           onclick: (e) => {
@@ -521,13 +764,15 @@ function agentRow(a) {
             stopAgent(a)
           },
         },
-        icon(ICONS.stop),
+        h("span", { class: "agent__square", "aria-hidden": "true" }),
       )
-    : h("span", { class: "agent__stop", "aria-hidden": "true" })
+    : h("span", { class: "agent__stop is-none", "aria-hidden": "true" })
   const facts = h(
     "p",
     { class: "agent__facts" },
     h("span", { class: "agent__state", text: stateWord(a.state) }),
+    h("span", { text: kindWord(a) }),
+    a.kind !== "member" && a.role ? h("code", { class: "agent__id", text: a.name || a.id }) : null,
     a.agent ? h("span", { text: a.agent }) : null,
     a.model ? h("span", { class: "agent__model", text: a.model }) : null,
     a.steps ? h("span", { text: t("%(steps)s steps, %(tools)s tools", { steps: a.steps, tools: a.tools }) }) : null,
@@ -547,8 +792,8 @@ function agentRow(a) {
     h(
       "summary",
       { class: "agent__row" },
-      h("span", { class: "agent__dot", "aria-hidden": "true" }),
+      h("span", { class: "agent__chev", "aria-hidden": "true" }, icon(ICONS.chevron)),
-      h("span", { class: "agent__what" }, h("span", { class: "agent__name", text: a.name || a.id }), a.role ? h("span", { class: "agent__role", text: a.role }) : null),
+      h("span", { class: "agent__what", title: label(a) }, h("bdi", { class: "agent__name", text: label(a) })),
       time,
       stopBtn,
     ),
@@ -556,12 +801,11 @@ function agentRow(a) {
       "div",
       { class: "agent__body" },
       facts,
+      a.kind === "member" && a.role ? h("p", { class: "agent__role", text: a.role }) : null,
       a.brief ? h("p", { class: "agent__brief", text: a.brief }) : null,
       step,
       a.text ? h("p", { class: "agent__text", text: a.text }) : null,
-      a.transcript
+      a.transcript ? h("button", { type: "button", class: "btn btn--sm", text: t("Open its conversation"), onclick: () => openAgent(a) }) : null,
-        ? h("button", { type: "button", class: "btn btn--sm", text: t("Open its conversation"), onclick: () => openAgent(a) })
-        : null,
     ),
   )
   el.style.setProperty("--depth", String(Math.max(0, Math.min(6, Number(a.depth) || 0))))
@@ -569,6 +813,22 @@ function agentRow(a) {
   return el
 }
 
+/**
+ * A task's line: a team's member by its name (it is how the room and the others call it); a
+ * subagent or a command by what it was given to do, its id only where there is nothing else.
+ */
+function label(a) {
+  if (a.kind === "member") return a.name || a.id
+  return a.role || a.brief || a.name || a.id
+}
+
+/** What a task is: a team's member, a subagent, or a background command. */
+function kindWord(a) {
+  if (a.kind === "member") return t("Team member")
+  if (a.type === "command") return t("Background command")
+  return t("Subagent")
+}
+
 async function stopAgent(a) {
   if (!chat) return
   if (a.kind === "member") {
@@ -602,6 +862,7 @@ function openAgent(a) {
 }
 
 function closeAgent() {
+  clearTimeout(subTimer)
   view = null
   render()
   $("activity-close").focus()
@@ -609,6 +870,7 @@ function closeAgent() {
 
 async function readAgent() {
   if (!chat || !view) return
+  if (view.kind === "sub") return readSub()
   const { kind, id } = view
   let r
   try {
@@ -634,7 +896,14 @@ async function readAgent() {
     a.role ? h("p", { class: "agent__role", text: a.role }) : null,
     a.stoppable ? h("button", { type: "button", class: "btn btn--sm", text: t("Stop %(name)s", { name: a.name || id }), onclick: () => stopAgent(a) }) : null,
   )
-  const turns = (r.transcript || []).map((turn) => {
+  const turns = turnsOf(r.transcript || [])
+  body.replaceChildren(head, h("div", { class: "convo" }, turns.length ? turns : h("p", { class: "activity__empty", text: t("Nothing yet.") })))
+  if (atEnd) body.scrollTop = body.scrollHeight
+}
+
+/** A conversation's turns, read-only: the person's (or the harness's) messages and the replies. */
+function turnsOf(list) {
+  return list.map((turn) => {
     if (turn.role === "user") {
       return h("article", { class: `msg msg--user convo__turn${turn.machine ? " is-machine" : ""}` }, h("div", { class: "msg__text", text: String(turn.text || "") }))
     }
@@ -661,14 +930,14 @@ async function readAgent() {
       turn.running ? h("p", { class: "msg__status muted", text: t("Writing…") }) : null,
     )
   })
-  body.replaceChildren(head, h("div", { class: "convo" }, turns.length ? turns : h("p", { class: "activity__empty", text: t("Nothing yet.") })))
-  if (atEnd) body.scrollTop = body.scrollHeight
 }
 
 // --- Start ------------------------------------------------------------------------------------------
 
-export function initActivity({ onLayout }) {
+export function initActivity({ onLayout, here, compacted }) {
   layout = onLayout || layout
+  onThisComputer = here || onThisComputer
+  onCompacted = compacted || onCompacted
   $("activity-close").replaceChildren(icon(ICONS.close))
   $("activity-close").addEventListener("click", () => showPanel(false))
   $("activity-back").replaceChildren(icon(ICONS.back))

ui/js/bridge.js

@@ -66,6 +66,61 @@ export async function follow(path, { onFrame, onEnd }) {
   }
 }
 
+// --- The terminal panel's socket -----------------------------------------------------------------
+//
+// Opened by Rust (`terminal_open`), which holds the token and the socket; the page gets the frames
+// as `terminal` events -- `output` (the shell's bytes, base64), `control` (the server's JSON), a last
+// `end` -- and sends keys, sizes and "close" through commands. As with a followed reply, a frame
+// can come before `terminal_open` has answered with its id: those wait.
+
+const terminals = new Map()
+const earlyTerminal = new Map()
+listen("terminal", (e) => {
+  const { id, kind, data } = e.payload
+  const f = terminals.get(id)
+  if (f) deliverTerminal(f, id, kind, data)
+  else {
+    if (!earlyTerminal.has(id)) earlyTerminal.set(id, [])
+    earlyTerminal.get(id).push([kind, data])
+  }
+})
+
+function deliverTerminal(f, id, kind, data) {
+  if (kind === "end") {
+    terminals.delete(id)
+    f.onEnd(data || {})
+  } else if (kind === "output") f.onOutput(data)
+  else if (kind === "control") f.onControl(data)
+}
+
+/**
+ * Open a shell: `target` is `{ chat }` or `{ device, directory }`; `onOutput(base64)`,
+ * `onControl(frame)`, and `onEnd({ code, reason, status, error })` once. The handle sends keys
+ * (`input`), a size (`resize`), and lets go (`close`; onEnd still comes).
+ */
+export async function openTerminal(target, cols, rows, { onOutput, onControl, onEnd }) {
+  const id = await invoke("terminal_open", {
+    chat: target.chat ?? null,
+    device: target.device ?? null,
+    directory: target.directory ?? null,
+    cols,
+    rows,
+  })
+  const f = { onOutput, onControl, onEnd }
+  terminals.set(id, f)
+  for (const [kind, data] of earlyTerminal.get(id) || []) {
+    if (!terminals.has(id)) break
+    deliverTerminal(f, id, kind, data)
+  }
+  earlyTerminal.delete(id)
+  return {
+    id,
+    input: (data) => invoke("terminal_input", { id, data }),
+    resize: (c, r) => invoke("terminal_resize", { id, cols: c, rows: r }),
+    close: () => invoke("terminal_close", { id }),
+  }
+}
+
 /** A web link, in the system's browser. */
 export function openLink(url) {
   return invoke("open_external", { url }).catch(() => {})

ui/js/chat.js

@@ -13,15 +13,17 @@ import { markdown } from "./markdown.js"
 import { openViewer, showKind } from "./viewer.js"
 import { readLocal } from "./explorer.js"
 import { menu, ask, showLine, popover, closePopover } from "./popup.js"
-import { avatar, agentsControl, teamPossible, levelsOf, levelWord, picture } from "./models.js"
+import { avatar, agentsControl, teamPossible, namedOnly, subagentsCap, noteSubagentsCap, SUBAGENTS_HINT, SUBAGENTS_OWN, levelsOf, levelWord, picture } from "./models.js"
 import { readyIds, sending, sent, discard } from "./attach.js"
-import { watchActivity, activityChanged, planList } from "./activity.js"
+import { watchActivity, activityChanged, planList, activityButton, treeChanged } from "./activity.js"
 import { agentBlock, agentReady, agentState, onAgentChange } from "./agent.js"
 import { folderName } from "./list.js"
+import { isProgress, progressLine, progressTicks } from "./progress.js"
+import { aimTerminal, terminalButton, terminalOpen, toggleTerminal } from "./terminal.js"
 
 const MODES = ["manual", "edit", "auto", "plan"]
 
-let ctx = null // set by initChat: { status(), thisDevices(), models(), me(), changed(chat), opened(chat), gone(chat) }
+let ctx = null // set by initChat: { status(), thisDevices(), devices(), models(), me(), changed(chat), opened(chat), gone(chat) }
 let open = null // { chat, live, more, first }
 
 export function openChatId() {
@@ -30,10 +32,80 @@ export function openChatId() {
 
 // --- Reading -----------------------------------------------------------------------------------
 
-function onThisComputer(chat) {
+export function onThisComputer(chat) {
   return !!chat?.device && ctx.thisDevices().includes(chat.device.token_id)
 }
 
+/** A linked device, as `GET /devices` lists it (with `terminal`: whether it opens one). */
+function deviceOf(id) {
+  return (ctx.devices?.() || []).find((d) => d.id === id) || null
+}
+
+/** The device a new agent chat goes to: this computer's, preferring the one linked right now. */
+function draftDevice() {
+  const mine = ctx.thisDevices()
+  return mine.find((id) => deviceOf(id)) || mine[0] || ""
+}
+
+/**
+ * What the Terminal panel is for, as the page stands: a new agent chat whose device and folder
+ * are chosen (the shell before the first message), an open agent chat on a device that opens
+ * terminals, or nothing.
+ */
+function terminalTarget() {
+  if (draft) {
+    if (draft.kind !== "agent" || !draft.dir) return null
+    const device = deviceOf(draftDevice())
+    if (!device?.terminal) return null
+    return { device: device.id, directory: draft.dir, name: device.name || t("This computer") }
+  }
+  const chat = open?.chat
+  if (chat?.kind !== "agent" || !chat.device) return null
+  const device = deviceOf(chat.device.token_id)
+  if (!device?.terminal) return null
+  return { chat: chat.id, device: device.id, directory: chat.device.directory, name: device.name || t("Terminal") }
+}
+
+// A new agent chat's subagent facts (spec 3.4.0 §14): `/models?kind=agent&device=` -- each model's
+// `subagents` for an agent chat, and `agent_subagents { offered, device_runs }` once. Read per
+// device, again when a new chat starts; a server before 3.4.0 says nothing (no setting, as before).
+const agentFacts = new Map() // device -> { at, models: Map(id -> subagents), offered, runs } | null
+
+async function readAgentFacts(device) {
+  if (!device) return null
+  const known = agentFacts.get(device)
+  if (known && Date.now() - known.at < 5000) return known
+  try {
+    const r = await get(`/models?kind=agent&device=${encodeURIComponent(device)}`)
+    if (!r.agent_subagents) {
+      agentFacts.set(device, { at: Date.now(), none: true })
+      return agentFacts.get(device)
+    }
+    noteSubagentsCap(r.subagents_cap)
+    const facts = { at: Date.now(), models: new Map((r.models || []).map((m) => [m.id, m.subagents])), offered: !!r.agent_subagents.offered, runs: r.agent_subagents.device_runs }
+    agentFacts.set(device, facts)
+    return facts
+  } catch {
+    return null
+  }
+}
+
+/** Whether a new agent chat on `model` is offered subagents: `{ cap, own }`, or null. */
+function agentOffer(facts, model) {
+  if (!facts || facts.none || !facts.offered || facts.runs === false) return null
+  const said = facts.models.get(model?.id)
+  if (said !== undefined && said !== null && !(typeof said === "object" ? said.offered : said)) return null
+  return { cap: subagentsCap(), own: namedOnly(said) }
+}
+
+/** Text into the message box, after what is there, for the person to read before sending. */
+export function intoComposer(text) {
+  const box = $("composer-text")
+  box.value = box.value.trim() ? `${box.value.replace(/\s*$/, "")}\n\n${text}` : text
+  box.dispatchEvent(new Event("input"))
+  box.focus()
+}
+
 function nearBottom() {
   const th = $("thread")
   return th.scrollHeight - th.scrollTop - th.clientHeight < 80
@@ -312,6 +384,24 @@ class Live {
     this.el = h("article", { class: "msg msg--assistant is-live", dataset: { id: messageId } }, this.steps, this.thinking, this.tail, this.jobs, this.status)
     this.done = false
     this.handle = null
+    this.progress = null // the stream's `progress` (spec 3.4.0), drawn in place of `status` when it comes
+    this.said = "" // the last `status` text, for a server that sends no progress
+    this.clock = 0
+  }
+
+  /** The status line: the progress object's line where the server sends one, else its text. */
+  drawStatus() {
+    const p = this.progress
+    this.status.textContent = isProgress(p) ? progressLine(p) : this.said || t("Writing…")
+    this.status.dataset.phase = isProgress(p) ? p.phase : ""
+    const ticks = progressTicks(p)
+    if (ticks && !this.clock) this.clock = setInterval(() => (this.done ? this.stopClock() : this.drawStatus()), 1000)
+    else if (!ticks) this.stopClock()
+  }
+
+  stopClock() {
+    clearInterval(this.clock)
+    this.clock = 0
   }
 
   async start() {
@@ -327,11 +417,18 @@ class Live {
 
   stop() {
     this.done = true
+    this.stopClock()
     this.handle?.close()
   }
 
   frame(event, data) {
     if (this.done) return
+    // Any frame that carries `progress` (the status frame does, from server 2.0.0-beta.8 on):
+    // the object, or null for nothing running -- then the status text again.
+    if (data && typeof data === "object" && Object.prototype.hasOwnProperty.call(data, "progress")) {
+      this.progress = isProgress(data.progress) ? data.progress : null
+      if (event !== "status") this.drawStatus()
+    }
     switch (event) {
       case "steps":
         renderSteps(this.steps, data.steps || [], this.chat)
@@ -341,11 +438,16 @@ class Live {
         this.tailed(data)
         break
       case "status":
-        this.status.textContent = data.text || t("Writing…")
+        this.said = data.text || ""
+        this.drawStatus()
         break
       case "ask":
         renderCards(this.chat, data.cards || [])
         break
+      case "tree":
+        // The subagent tree as it changed (spec 3.4.0): the activity panel draws it.
+        treeChanged(this.chat.id, data.tree)
+        break
       case "jobs":
         this.jobs.hidden = !(data.jobs || []).length
         append(
@@ -360,6 +462,7 @@ class Live {
         break
       case "done":
         this.done = true
+        this.stopClock()
         this.el.replaceWith(assistantMessage(data.message, this.chat))
         renderCards(this.chat, [])
         if (data.chat?.title) setTitle(data.chat.title)
@@ -385,6 +488,7 @@ class Live {
   end(err) {
     if (this.done) return
     this.done = true
+    this.stopClock()
     if (open?.live === this) open.live = null
     setRunning(false)
     if (err) {
@@ -417,10 +521,32 @@ function follower(chat, messageId) {
 
 // --- Cards ----------------------------------------------------------------------------------------
 
+/**
+ * The cards waiting for an answer, as the stream's `ask` frame lists them. A card already on the
+ * screen is kept as it is -- the server sends the frame again whenever anything in it changes, its
+ * countdown included, and drawing it again would take away a command being corrected or an option
+ * already picked. New ones are added in the server's order; the ones no longer listed go.
+ */
 function renderCards(chat, cards) {
-  const box = clear($("cards"))
+  const box = $("cards")
-  box.hidden = !cards.length
+  const ids = cards.map((c) => String(c.id))
-  for (const card of cards) box.append(cardOf(chat, card))
+  for (const el of [...box.children]) if (el.dataset.chat !== chat.id || !ids.includes(el.dataset.card)) el.remove()
+  cards.forEach((card, i) => {
+    let el = [...box.children].find((x) => x.dataset.card === String(card.id))
+    if (!el) {
+      try {
+        el = cardOf(chat, card)
+      } catch (err) {
+        // Never an empty box under a status that says the reply waits: say what went wrong.
+        console.error("card", card, err)
+        el = h("section", { class: "ask" }, h("p", { class: "msg__note is-error", text: String(err) }))
+      }
+      el.dataset.card = String(card.id)
+      el.dataset.chat = chat.id
+    }
+    if (box.children[i] !== el) box.insertBefore(el, box.children[i] || null)
+  })
+  box.hidden = !box.children.length
 }
 
 function cardOf(chat, card) {
@@ -430,24 +556,44 @@ function cardOf(chat, card) {
   if (card.why) el.append(h("p", { class: "muted", text: card.why }))
   for (const item of card.items) {
     const part = h("div", { class: "ask__item" })
+    // A background subagent asks, not the conversation's own turn: said first (`job`).
+    if (item.job) part.append(h("p", { class: "ask__header", text: t("%(job)s asks:", { job: item.job }) }))
+    // A subagent of the tree asks, for a call nobody in the tree may allow (spec 3.4.0, §7).
+    if (item.subagent) {
+      const sub = item.subagent
+      part.append(h("p", { class: "ask__header ask__subagent", text: t("Subagent %(name)s (%(id)s), %(position)s, asks:", { name: sub.name || sub.id, id: sub.id, position: sub.position || "—" }) }))
+    }
     if (item.header) part.append(h("p", { class: "ask__header", text: item.header }))
+    let detailShown = false
     if (item.kind === "approval") {
       part.append(h("p", { class: "ask__title", text: item.tool ? t("The agent asks to run %(tool)s:", { tool: item.tool }) : t("The agent asks:") }))
       if (item.editable) {
-        const edit = h("textarea", { class: "ask__edit", rows: "2", "aria-label": t("Command"), value: item.title })
+        // A command, correctable before it runs. The server says it in `detail` and puts a fixed
+        // "Run a command" in the title (services/device_link.py); a corrected command goes back as
+        // this item's text, and only when it differs -- an untouched one is not an edit.
+        const command = item.detail || item.title
+        detailShown = true
+        const edit = h("textarea", { class: "ask__edit", rows: "2", spellcheck: "false", "aria-label": t("Command"), value: command })
         edit.addEventListener("input", () => {
-          if (edit.value !== item.title) answers.texts[item.key] = edit.value
+          if (edit.value !== command) answers.texts[item.key] = edit.value
           else delete answers.texts[item.key]
         })
         part.append(edit)
-      } else part.append(h("pre", { class: "call__pre", text: item.title }))
+      } else {
+        // What it wants to do (a tool's own title), then its input as the device sent it.
+        part.append(h("p", { class: "ask__what", text: item.title }))
+        if (item.detail) part.append(h("pre", { class: "call__pre", text: item.detail }))
+        detailShown = true
+      }
     } else if (item.kind === "plan" && item.plan) {
       part.append(h("p", { class: "ask__title", text: item.plan.title || item.title }), planList(item.plan))
     } else {
       part.append(h("p", { class: "ask__title", text: item.title }))
     }
-    if (item.detail) part.append(h("p", { class: "muted", text: item.detail }))
+    if (item.detail && !detailShown) part.append(h("p", { class: "muted", text: item.detail }))
-    if (item.purpose) part.append(h("p", { class: "muted", text: item.purpose }))
+    // The model's account of what it is for, attributed to it; then the server's reason for asking.
+    if (item.purpose) part.append(h("p", { class: "muted", text: t("It says: %(purpose)s", { purpose: item.purpose }) }))
+    if (item.reason) part.append(h("p", { class: "muted ask__reason", text: item.reason }))
     if (item.options?.length) {
       const picked = new Set()
       const opts = h("div", { class: "ask__options", role: item.multiple ? "group" : "radiogroup" })
@@ -506,12 +652,11 @@ function cardOf(chat, card) {
   const actions = h("div", { class: "ask__actions" })
   if (decides) {
     const isPlan = card.items.some((i) => i.kind === "plan")
-    append(
+    append(actions, [
-      actions,
       h("button", { type: "button", class: "btn btn--primary", text: isPlan ? t("Carry it out") : t("Allow"), onclick: () => send("allow") }),
       isPlan ? null : h("button", { type: "button", class: "btn", text: t("Always allow"), onclick: () => send("allow_always") }),
       h("button", { type: "button", class: "btn", text: isPlan ? t("Not yet") : t("Deny"), onclick: () => send("deny") }),
-    )
+    ])
   } else {
     actions.append(h("button", { type: "button", class: "btn btn--primary", text: t("Answer"), onclick: () => send("") }))
   }
@@ -542,16 +687,11 @@ function renderHead(chat) {
         h("span", { class: "place__where", text: here ? t("This computer") : t("Another device") }),
       ),
     )
-    if (chat.device_session_id) {
+    // The Terminal panel where the chat's device opens one (`GET /devices` says so per device);
-      tools.append(
+    // "Continue in a terminal" stays in the chat's menu.
-        h(
+    if (deviceOf(chat.device.token_id)?.terminal) tools.append(terminalButton())
-          "button",
-          { type: "button", class: "tool", "aria-label": t("Continue in a terminal"), title: t("Continue in a terminal"), onclick: () => continueInTerminal(open?.chat || chat) },
-          icon(ICONS.terminal),
-        ),
-      )
-    }
   }
+  tools.append(activityButton())
   tools.append(h("button", { type: "button", class: "tool", "aria-label": t("More"), title: t("More"), onclick: (e) => chatMenu(e.currentTarget, open?.chat || chat) }, icon(ICONS.more)))
 }
 
@@ -564,13 +704,23 @@ async function chatModels(chat) {
   }
 }
 
-/** Whether this chat may have a team (`GET /chats/{chat}/team`): only a chat on the server. */
+/**
+ * Whether this chat is offered subagents: `{ cap, own }` (the instance's most; whether only a model
+ * the person names could serve them), or null. The chat's own `subagents` where the server gives
+ * it (2.0.0-beta.7), else its team's word (`GET /chats/{chat}/team`) -- only a chat on the server.
+ */
 async function teamOffered(chat) {
-  if (chat.kind !== "chat") return false
+  const said = chat.subagents
+  if (said !== undefined && said !== null) {
+    const offered = typeof said === "object" ? said.offered : said
+    return offered ? { cap: subagentsCap(said), own: namedOnly(said) } : null
+  }
+  if (chat.kind !== "chat") return null
   try {
-    return !!(await get(`/chats/${encodeURIComponent(chat.id)}/team`)).team?.offered
+    const team = (await get(`/chats/${encodeURIComponent(chat.id)}/team`)).team
+    return team?.offered ? { cap: subagentsCap(), own: false } : null
   } catch {
-    return false
+    return null
   }
 }
 
@@ -640,26 +790,31 @@ function effortChip(model, value, onChange) {
   })
 }
 
-/** The team's size: a chip saying it, opening the two fields. `value` { min, max } or empty. */
+/**
-function agentsChip(value, onChange) {
+ * The subagents a chat may run: a chip saying Min–Max, opening one range with two handles.
+ * `value` { min, max } or empty; `cap` the instance's most, where the server says it.
+ */
+function agentsChip(value, onChange, { cap, own } = {}) {
   const said = value?.min || value?.max ? `${value.min ?? "…"}–${value.max ?? "…"}` : t("default")
   return chip({
     ico: ICONS.users,
-    label: `${t("Agents")} ${said}`,
+    label: `${t("Subagents")} ${said}`,
-    title: t("Its own agent included. Above one, it can add members that do the work while it leads them."),
+    title: own ? SUBAGENTS_OWN() : SUBAGENTS_HINT(),
+    cls: `cchip--subagents${own ? " is-warn" : ""}`,
     onclick: (e) => {
-      const fields = agentsControl({ value, id: "agents-min" })
+      const fields = agentsControl({ value, id: "agents-min", cap })
       popover(
         e.currentTarget,
         h(
           "div",
           { class: "pop__body" },
-          h("label", { class: "pop__title", for: "agents-min", text: t("Agents") }),
+          h("p", { class: "pop__title", text: t("Subagents") }),
           fields,
-          h("p", { class: "hint", text: t("Its own agent included. Above one, it can add members that do the work while it leads them.") }),
+          own ? h("p", { class: "hint is-warn", text: SUBAGENTS_OWN() }) : null,
+          h("p", { class: "hint", text: SUBAGENTS_HINT() }),
           h("div", { class: "pop__actions" }, h("button", { type: "button", class: "btn btn--sm btn--primary", text: t("Save"), onclick: () => (closePopover(), onChange(fields.value)) })),
         ),
-        { up: true, label: t("Agents") },
+        { up: true, label: t("Subagents") },
       )
     },
   })
@@ -699,17 +854,38 @@ async function renderChips() {
       parts.push(folderChip(roots, draft.dir, (d) => ((draft.dir = d), renderChips())))
       parts.push(modeChip(draft.mode, (m) => ((draft.mode = m), renderChips())))
     }
+    const shell = terminalTarget()
+    aimTerminal(shell)
     parts.push(modelChip(models, draft.model, (id) => {
       draft.model = id
       draft.effort = levelsOf(models.find((m) => m.id === id)).start
       renderChips()
     }, mayPick))
     if (model) parts.push(effortChip(model, draft.effort, (l) => ((draft.effort = l), renderChips())))
-    if (draft.kind === "chat" && teamPossible(model)) parts.push(agentsChip(draft.agents, (a) => ((draft.agents = a), renderChips())))
+    if (draft.kind === "chat" && teamPossible(model)) parts.push(agentsChip(draft.agents, (a) => ((draft.agents = a), renderChips()), { cap: subagentsCap(), own: namedOnly(model?.subagents) }))
+    // An agent chat's subagents run on its device: offered as the server says for that device.
+    const device = draft.kind === "agent" ? draftDevice() : ""
+    const known = device ? agentFacts.get(device) : null
+    const offer = device ? agentOffer(known, model) : null
+    if (offer) parts.push(agentsChip(draft.agents, (a) => ((draft.agents = a), renderChips()), offer))
+    if (device && (!known || Date.now() - known.at > 5000)) {
+      readAgentFacts(device).then((f) => {
+        if (f && drawing === chipsDrawn && draft?.kind === "agent") renderChips()
+      })
+    }
+    // Last, so the chat's own settings come first where the line scrolls.
+    if (shell) {
+      const b = chip({ ico: ICONS.terminal, label: t("Terminal"), menuish: false, cls: "cchip--terminal", onclick: toggleTerminal })
+      b.dataset.terminalToggle = ""
+      b.setAttribute("aria-pressed", String(terminalOpen()))
+      b.setAttribute("aria-controls", "terminal")
+      parts.push(b)
+    }
     box.replaceChildren(...parts.filter(Boolean))
     return
   }
   const chat = open?.chat
+  aimTerminal(terminalTarget())
   if (!chat?.kind) return box.replaceChildren()
   const save = async (body) => {
     try {
@@ -732,8 +908,9 @@ async function renderChips() {
   if (chat.model) parts.push(modelChip(models, chat.model, (id) => save({ model: id }), mayPick))
   parts.push(effortChip(current, chat.effort || levelsOf(current).start, (l) => save({ effort: l })))
   box.replaceChildren(...parts.filter(Boolean))
-  if (!(await teamOffered(chat)) || drawing !== chipsDrawn || open?.chat?.id !== chat.id) return
+  const offer = await teamOffered(chat)
-  box.append(agentsChip(open.chat.agents || chat.agents, (a) => save({ agents: a })))
+  if (!offer || drawing !== chipsDrawn || open?.chat?.id !== chat.id) return
+  box.append(agentsChip(open.chat.agents || chat.agents, (a) => save({ agents: a }), offer))
 }
 
 /**
@@ -756,6 +933,10 @@ export function statusChanged() {
   if (draft) {
     renderChips()
     renderHello()
+  } else if (open?.chat?.kind) {
+    // The devices were read again: whether this chat's opens a terminal may have changed.
+    renderHead(open.chat)
+    aimTerminal(terminalTarget())
   }
 }
 
@@ -943,6 +1124,18 @@ export function draftKind() {
   return draft?.kind || lastKind
 }
 
+/** `/models` was read again: a new chat being written is drawn with what it says now. */
+export function modelsChanged() {
+  if (!draft) return
+  const models = ctx.models()
+  if (!models.some((m) => m.id === draft.model)) {
+    const start = models.find((m) => m.default) || models[0]
+    draft.model = start?.id || ""
+    draft.effort = levelsOf(start).start
+  }
+  renderChips()
+}
+
 /** The greeting of a new chat, and the choice of what it is. */
 function renderHello() {
   const box = $("new-chat")
@@ -1027,7 +1220,7 @@ async function createChat(text, attachments) {
   const agents = draft.agents
   if (agents && (agents.min || agents.max)) body.agents = agents
   if (draft.kind === "agent") {
-    const device = ctx.thisDevices()[0]
+    const device = draftDevice()
     if (!device || !agentReady(ctx.status())) {
       renderNotice()
       throw { detail: agentState(ctx.status()).sentence }

ui/js/dom.js

@@ -27,7 +27,8 @@ export function h(tag, attrs, ...children) {
   return el
 }
 
-export function append(el, children) {
+/** append(el, child, [more], …): children one by one or in arrays, nested as deep as they come. */
+export function append(el, ...children) {
   for (const c of children.flat(Infinity)) {
     if (c === null || c === undefined || c === false) continue
     el.append(c instanceof Node ? c : document.createTextNode(String(c)))
@@ -75,12 +76,14 @@ export const ICONS = {
   max: "M4 4h8v8H4zM5.5 5.5v5h5v-5z",
   restore: "M5.5 3h7.5v7.5h-2.5V13H3V5.5h2.5zM7 4.5v1h3.5V9h1V4.5zM4.5 7v4.5H9V7z",
   chevronDown: "M3.5 6L8 10.5 12.5 6 11.4 4.9 8 8.3 4.6 4.9z",
+  chevronUp: "M3.5 10L8 5.5 12.5 10 11.4 11.1 8 7.7 4.6 11.1z",
   more: "M2.5 6.75h2.5v2.5H2.5zM6.75 6.75h2.5v2.5h-2.5zM11 6.75h2.5v2.5H11z",
   edit: "M10.75 1.75l3.5 3.5-8.5 8.5h-3.5v-3.5zM3.75 10.87v1.38h1.38l6.62-6.62-1.38-1.38z",
   copy: "M5 1.5h8.5V11H5zM6.5 3v6.5H12V3zM2.5 5H4v8.5h6.5V15h-8z",
   attach: "M7.25 2.5h1.5v4.75h4.75v1.5H8.75v4.75h-1.5V8.75H2.5v-1.5h4.75z",
   bolt: "M9.5 1L3 9.25h4.25L6.5 15 13 6.75H8.75z",
   stop: "M4.5 4.5h7v7h-7z",
+  tasks: "M2 3.25h2v2H2zM5.5 3.5H14V5H5.5zM2 7h2v2H2zM5.5 7.25H14v1.5H5.5zM2 10.75h2v2H2zM5.5 11H14v1.5H5.5z",
   back: "M10 3.5L5.5 8 10 12.5 11.1 11.4 7.7 8 11.1 4.6z",
   terminal: "M1.5 2.5h13v11h-13zM3 4v8h10V4zM4.2 5.4l2.6 2.6-2.6 2.6-1-1L4.8 8 3.2 6.4zM7.5 10h4v1.5h-4z",
   search: "M7 2a5 5 0 013.95 8.07l3.24 3.24-1.06 1.06-3.24-3.24A5 5 0 117 2zm0 1.5a3.5 3.5 0 100 7 3.5 3.5 0 000-7z",

ui/js/main.js

@@ -12,13 +12,15 @@ import { renderSettings, initSettings, scrollToSection } from "./settings.js"
 import { initViewer, onViewer, closeViewer } from "./viewer.js"
 import { initExplorer, showLocal, showFilesTab } from "./explorer.js"
 import { initList, refreshList, markCurrent } from "./list.js"
-import { initChat, openChat, newChat, closeChat, openChatId, chatMenu, statusChanged, draftKind } from "./chat.js"
+import { initChat, openChat, newChat, closeChat, openChatId, chatMenu, statusChanged, draftKind, onThisComputer, modelsChanged, intoComposer } from "./chat.js"
+import { initTerminal, showTerminal, terminalOpen, terminalAvailable, toggleTerminal, terminalThemeChanged } from "./terminal.js"
 import { initAttach, allowAttach } from "./attach.js"
 import { initPalette, openPalette } from "./palette.js"
 import { initActivity, showPanel, togglePanel, panelOpen } from "./activity.js"
 import { initAgent, link } from "./agent.js"
 import { initScheduled, showScheduled, scheduledCommands } from "./scheduled.js"
 import { closePopover } from "./popup.js"
+import { noteSubagentsCap } from "./models.js"
 
 let status = null
 let view = "" // "setup" | "app"
@@ -66,6 +68,7 @@ function applyTheme(name) {
   const root = document.documentElement
   root.dataset.theme = name
   root.dataset.base = name
+  terminalThemeChanged()
 }
 
 // --- Views ------------------------------------------------------------------------------------
@@ -116,11 +119,17 @@ function stopApp() {
   closeViewer()
 }
 
-async function loadModels() {
+let modelsAt = 0 // when /models was last read
+
+/** The account's models, with what a new chat on each is offered. `keep`: a failed read keeps the last list. */
+async function loadModels({ keep = false } = {}) {
   try {
-    models = (await get("/models")).models || []
+    const r = await get("/models")
+    models = r.models || []
+    noteSubagentsCap(r.subagents_cap)
+    modelsAt = Date.now()
   } catch {
-    models = []
+    if (!keep) models = []
   }
 }
 
@@ -155,7 +164,10 @@ function showPage(name) {
   $("page-scheduled").hidden = name !== "scheduled"
   $("settings").hidden = name !== "settings"
   for (const b of document.querySelectorAll(".navrow[data-view]")) b.setAttribute("aria-current", String(b.dataset.view === name))
-  if (name !== "chat") showPanel(false)
+  if (name !== "chat") {
+    showPanel(false)
+    showTerminal(false)
+  }
   if (narrow()) sideOpen(false)
 }
 
@@ -174,6 +186,10 @@ function startNew(kind) {
   newChat(kind || draftKind())
   markCurrent()
   localPane()
+  // What a new chat is offered (its models, Subagents) is the server's word now, not at sign-in:
+  // an administrator may have switched subagents on or added a model while the app stayed open in
+  // the tray. Drawn at once from the last list, then again from the fresh one.
+  if (started && Date.now() - modelsAt > 5000) loadModels({ keep: true }).then(modelsChanged)
 }
 
 /** Open a chat, from the list or the palette. */
@@ -210,6 +226,7 @@ function commands() {
     { label: t("Settings"), run: () => showSettings(true) },
   ]
   if (openChatId()) out.push({ label: panelOpen() ? t("Close the activity panel") : t("Open the activity panel"), run: togglePanel })
+  if (terminalAvailable()) out.push({ label: terminalOpen() ? t("Close terminal") : t("Terminal"), run: toggleTerminal })
   if (status?.linked) out.push({ label: status.agent_wanted ? t("Stop the agent") : t("Start the agent"), run: () => invoke("set_agent", { on: !status.agent_wanted }) })
   out.push(
     { label: status?.linked ? t("Link again") : t("Link this computer"), run: link },
@@ -254,6 +271,7 @@ async function boot() {
   initChat({
     status: () => status,
     thisDevices,
+    devices: () => devices,
     models: () => models,
     me: () => me,
     changed: () => refreshList(),
@@ -278,10 +296,22 @@ async function boot() {
   onViewer((on) => {
     $("shell").dataset.viewer = String(on)
   })
+  // One panel column: the activity panel and the terminal take turns in it (the shell stays open).
+  initTerminal({
+    onLayout: (on) => {
+      $("shell").dataset.terminal = String(on)
+      if (on && panelOpen()) showPanel(false)
+    },
+    composer: intoComposer,
+  })
   initActivity({
     onLayout: (on) => {
       $("shell").dataset.activity = String(on)
+      if (on && terminalOpen()) showTerminal(false)
     },
+    here: onThisComputer,
+    // The conversation as it now stands: the summary's divider, read again, if it is still open.
+    compacted: (id) => openChatId() === id && openChat(id),
   })
 
   $("new-chat-btn").prepend(icon(ICONS.plus))
@@ -339,7 +369,7 @@ async function boot() {
     renderBar(status)
     route()
     // The agent came up or went: which devices are linked changes with it.
-    if (started && before && before.agent.state !== status.agent.state) loadDevices().then(refreshList)
+    if (started && before && before.agent.state !== status.agent.state) loadDevices().then(() => (refreshList(), statusChanged()))
   })
   listen("open-settings", () => {
     if (view === "app") showSettings(true)

ui/js/models.js

@@ -123,7 +123,7 @@ export function modelPicker({ id, models, value, onChange, label }) {
   return btn
 }
 
-// --- How hard a model thinks, and how many agents a chat may have --------------------------------
+// --- How hard a model thinks, and how many subagents a chat may have --------------------------------
 //
 // The levels are the harness spec's (3.2.0, `models/schema.json`), as /api/v2/models gives them:
 // `levels` in offering order (`off` first where the model can switch thinking off, then its
@@ -193,69 +193,120 @@ export function effortControl({ model, value, onChange, id, compact }) {
 }
 
 /**
- * Whether a new chat on `model` may have a team: a chat on the server (an agent chat's team runs
+ * Whether a new chat on `model` is offered subagents: the model's `subagents.offered` in `/models`
- * on its device), a model that calls tools and serves two requests at once (`parallel` null is a
+ * (server 2.0.0-beta.7: subagents on for the model and the instance, and at least one model could
- * hosted API with no stated limit). The server decides again; this only keeps the fields away
+ * serve as one -- the model itself with room for two requests at once, or a usable model outside
- * where they could never apply. An open chat asks the server (`GET /chats/{chat}/team`).
+ * its one-model-at-a-time server). A server that does not say it: a model that calls tools and
+ * serves two requests at once (`parallel` null is a hosted API with no stated limit). The server
+ * decides again; this only keeps the chip away where it could never apply. An open chat has its
+ * own `subagents` (chat.js).
  */
 export function teamPossible(model) {
+  const said = model?.subagents
+  if (said !== undefined && said !== null) return typeof said === "object" ? !!said.offered : !!said
   return !!model?.tools && (model.parallel === null || model.parallel === undefined || model.parallel >= 2)
 }
 
+/** Whether only a model the person names in the chat could serve its subagents (`named_only`). */
+export function namedOnly(said) {
+  return !!(said && typeof said === "object" && said.named_only)
+}
+
+// The instance's most, as `/models` said it last (`subagents_cap`); 32 where it says nothing.
+export const SUBAGENTS_CAP = 32
+let knownCap = null
+export function noteSubagentsCap(n) {
+  knownCap = Number(n) >= 1 ? Number(n) : null
+}
+/** The most subagents a chat may have: the first of `sources` that says it (`cap`), else the instance's. */
+export function subagentsCap(...sources) {
+  for (const s of sources) {
+    const n = Number(s?.cap)
+    if (Number.isFinite(n) && n >= 1) return n
+  }
+  return knownCap || SUBAGENTS_CAP
+}
+// The numbers count the chat's own agent: 1 is the agent alone.
+export const SUBAGENTS_HINT = () => t("The numbers count this chat's own agent: 1 is the agent alone, with no subagents; Max 4 is the agent and up to 3 subagents. A subagent runs on the chat's model, or on another only when you name it in the chat.")
+// The spec's `subagents.setting.own`: beside the setting where the chat's model cannot serve them.
+export const SUBAGENTS_OWN = () => t("This model serves one request at a time: subagents need another model, on another server, named in the chat.")
+
+// A new chat's setting as the harness spec seeds it (`loop.json` agents): 1 to 4.
+export const AGENTS_DEFAULT = { min: 1, max: 4 }
+
 /**
- * The team's size: the fewest and the most agents, the chat's own agent included, as two number
+ * The agents a chat may run, its own agent counted: one range from 1 to `cap` with two handles,
- * fields on one shared track. `value` is { min, max } (empty fields for a new chat: the person's
+ * Min and Max -- two range inputs overlaid on one track, the span between them filled. Each is a
- * default for new chats then stands); `onChange({ min, max })` once a field is changed, with the
+ * real slider (arrow keys, Home/End, Page Up/Down), and Min never passes Max. `value` is
- * empty one left out. The server checks the pair and says the bounds when it refuses one.
+ * { min, max } (empty for a new chat: the spec's defaults are shown, and left to the server
+ * unless a handle moves); `el.value` the pair. The server checks it again.
  */
-export function agentsControl({ value, onChange, id, compact }) {
+export function agentsControl({ value, onChange, id, cap = SUBAGENTS_CAP }) {
-  const field = (which, v, label) =>
+  const top = Math.max(1, Number(cap) || SUBAGENTS_CAP)
+  const clamp = (n, d) => Math.min(top, Math.max(1, Number.isFinite(Number(n)) && n !== null && n !== "" ? Number(n) : d))
+  let given = value || {}
+  let touched = false
+  const handle = (which, label) =>
     h("input", {
-      type: "number",
+      type: "range",
-      class: "agents__input",
+      class: `range2__input range2__input--${which}`,
       id: which === "min" ? id : null,
       min: "1",
+      max: String(top),
       step: "1",
-      inputmode: "numeric",
-      value: v === undefined || v === null ? "" : String(v),
-      placeholder: compact ? null : t("default"),
       "aria-label": label,
       title: label,
     })
-  const low = field("min", value?.min, t("Fewest"))
+  const low = handle("min", t("Min"))
-  const high = field("max", value?.max, t("Most"))
+  const high = handle("max", t("Max"))
-  const bounds = () => {
+  const said = h("output", { class: "range2__said", "aria-live": "polite", for: id || null })
-    high.min = low.value || "1"
+  const fill = h("span", { class: "range2__fill", "aria-hidden": "true" })
-    if (high.value) low.max = high.value
+  const box = h("span", { class: "range2__box" }, h("span", { class: "range2__track", "aria-hidden": "true" }, fill), low, high)
-    else low.removeAttribute("max")
+  const draw = () => {
+    const lo = Number(low.value)
+    const hi = Number(high.value)
+    const at = (n) => (top > 1 ? ((n - 1) / (top - 1)) * 100 : 0)
+    box.style.setProperty("--lo", `${at(lo)}%`)
+    box.style.setProperty("--hi", `${at(hi)}%`)
+    // Where the two meet, the one that can still move is on top: Min at the far end, else Max.
+    low.classList.toggle("is-top", lo === hi && hi === top)
+    low.setAttribute("aria-valuetext", `${t("Min")} ${lo}`)
+    high.setAttribute("aria-valuetext", `${t("Max")} ${hi}`)
+    said.textContent = `${t("Min")} ${lo} · ${t("Max")} ${hi}`
   }
-  bounds()
+  const set = (v) => {
-  const read = () => {
+    const hi = clamp(v?.max, Math.min(AGENTS_DEFAULT.max, top))
-    const out = {}
+    const lo = Math.min(hi, clamp(v?.min, AGENTS_DEFAULT.min))
-    if (low.value) out.min = Number(low.value)
+    high.value = String(hi)
-    if (high.value) out.max = Number(high.value)
+    low.value = String(lo)
-    return out
+    draw()
-  }
-  for (const input of [low, high]) {
-    input.addEventListener("change", () => {
-      bounds()
-      onChange?.(read())
-    })
   }
+  low.addEventListener("input", () => {
+    if (Number(low.value) > Number(high.value)) low.value = high.value
+    touched = true
+    draw()
+  })
+  high.addEventListener("input", () => {
+    if (Number(high.value) < Number(low.value)) high.value = low.value
+    touched = true
+    draw()
+  })
+  for (const input of [low, high]) input.addEventListener("change", () => onChange?.(read()))
+  const read = () => (touched || given.min || given.max ? { min: Number(low.value), max: Number(high.value) } : {})
+  set(given)
   const el = h(
     "span",
-    { class: `agents${compact ? " agents--compact" : ""}`, role: "group", "aria-label": t("Agents"), title: t("Its own agent included. Above one, it can add members that do the work while it leads them.") },
+    { class: "range2", role: "group", "aria-label": t("Subagents") },
-    compact ? h("span", { class: "agents__label", text: t("Agents") }) : null,
+    said,
-    low,
+    box,
-    h("span", { class: "agents__dash", "aria-hidden": "true", text: "–" }),
+    h("span", { class: "range2__ends", "aria-hidden": "true" }, h("span", { text: "1" }), h("span", { text: String(top) })),
-    high,
   )
   Object.defineProperty(el, "value", {
     get: read,
     set: (v) => {
-      low.value = v?.min ?? ""
+      given = v || {}
-      high.value = v?.max ?? ""
+      touched = false
-      bounds()
+      set(given)
     },
   })
   return el

ui/js/progress.js

@@ -0,0 +1,134 @@
+// What a session is doing, as one line: the harness's `progress` object (spec 3.4.0,
+// harness/progress.md in the LLeMbas CLI) drawn the way every client draws it --
+// `{{text}} · {{elapsed}}`, with the shared texts (`loop.json` progress.texts) as this page's
+// sentences. The server sends the object beside the old `status` text in a chat's stream; where it
+// sends none (an older server), the status text is drawn as before.
+//
+// The texts are the spec's, word for word with their `{{field}}`s: the server's catalogues
+// translate them under the same keys (scripts/gen-i18n.py takes them from there).
+
+/** The thirteen phases: a text for each, and a variant where a field may be missing. */
+const TEXTS = {
+  queued: "Queued",
+  "queued.ahead": "Queued: {{ahead}} ahead",
+  waiting_for_model: "Waiting for {{model}}: {{reason}}",
+  loading_model: "Loading {{model}}",
+  reading_prompt: "Reading the prompt: {{done}} / {{total}} tokens ({{percent}}%)",
+  "reading_prompt.bare": "Reading the prompt",
+  thinking: "Thinking",
+  writing: "Writing: {{rate}} tok/s",
+  "writing.bare": "Writing",
+  running_tool: "Running {{tool}}: {{what}}",
+  "running_tool.bare": "Running {{tool}}",
+  waiting_for_person: "Waiting for you: {{what}}",
+  waiting_for_subagents: "Waiting for subagents: {{running}} running",
+  compacting: "Compacting the conversation",
+  retrying: "Retrying in {{in}} s: {{reason}}",
+  stopped: "Stopped",
+  "stopped.reason": "Stopped: {{reason}}",
+  failed: "Failed: {{reason}}",
+}
+
+/** The words a field stands for. */
+const REASONS = {
+  busy: "it serves one request at a time and is answering another",
+  slots: "all its slots are in use",
+  group: "its server holds another model now",
+}
+const WHAT = {
+  approval: "allow {{tool}}?",
+  "approval.bare": "an approval",
+  question: "a question",
+  plan: "the plan",
+}
+const STOPPED = { person: "by you", steps: "step limit", budget: "budget" }
+
+/** Every sentence above, for scripts/gen-i18n.py to find (it reads `t("…")` calls by hand). */
+export const PROGRESS_SENTENCES = [...Object.values(TEXTS), ...Object.values(REASONS), ...Object.values(WHAT), ...Object.values(STOPPED)]
+
+const NO_ELAPSED = ["stopped", "failed"]
+
+const has = (p, k) => p[k] !== undefined && p[k] !== null && p[k] !== ""
+
+/** A sentence in the page's language (the spec's text is the key, as in the server's
+ * catalogues), its `{{field}}`s filled -- after the lookup, so a field's own text is never read
+ * as a template. */
+function say(text, fields = {}) {
+  return t(text).replace(/\{\{(\w+)\}\}/g, (all, name) => (Object.prototype.hasOwnProperty.call(fields, name) ? String(fields[name]) : all))
+}
+
+/** An integer as it is; a rate with one decimal, in the page's language (12.5, 12,5). */
+function rate(n) {
+  try {
+    return new Intl.NumberFormat(document.documentElement.lang || "en", { useGrouping: false, minimumFractionDigits: 1, maximumFractionDigits: 1 }).format(n)
+  } catch {
+    return Number(n).toFixed(1)
+  }
+}
+const int = (n) => String(Math.trunc(Number(n) || 0))
+
+/** Whole seconds, as a ticking clock writes them: 5s, 1m 05s, 1h 02m. */
+export function elapsed(ms) {
+  const total = Math.max(0, Math.floor((Number(ms) || 0) / 1000))
+  if (total < 60) return `${total}s`
+  const minutes = Math.floor(total / 60)
+  if (minutes < 60) return `${minutes}m ${String(total % 60).padStart(2, "0")}s`
+  return `${Math.floor(minutes / 60)}h ${String(minutes % 60).padStart(2, "0")}m`
+}
+
+/** Whether `p` is a progress object this page can draw (one of the thirteen phases). */
+export function isProgress(p) {
+  return !!p && typeof p === "object" && Object.prototype.hasOwnProperty.call(TEXTS, p.phase)
+}
+
+/** The phase's text, its fields filled; "" for nothing this page knows. */
+export function progressText(p, now = Date.now()) {
+  if (!isProgress(p)) return ""
+  switch (p.phase) {
+    case "queued":
+      return has(p, "ahead") ? say(TEXTS["queued.ahead"], { ahead: int(p.ahead) }) : say(TEXTS.queued)
+    case "waiting_for_model":
+      return say(TEXTS.waiting_for_model, { model: p.model || "", reason: REASONS[p.reason] ? say(REASONS[p.reason]) : p.reason || "" })
+    case "loading_model":
+      return say(TEXTS.loading_model, { model: p.model || "" })
+    case "reading_prompt":
+      if (!has(p, "done") || !(Number(p.total) > 0)) return say(TEXTS["reading_prompt.bare"])
+      return say(TEXTS.reading_prompt, { done: int(p.done), total: int(p.total), percent: int(Math.floor((Number(p.done) * 100) / Number(p.total))) })
+    case "writing":
+      return has(p, "rate") ? say(TEXTS.writing, { rate: rate(p.rate) }) : say(TEXTS["writing.bare"])
+    case "running_tool":
+      return has(p, "what") ? say(TEXTS.running_tool, { tool: p.tool || "", what: p.what }) : say(TEXTS["running_tool.bare"], { tool: p.tool || "" })
+    case "waiting_for_person": {
+      const what = p.what === "approval" ? (has(p, "tool") ? say(WHAT.approval, { tool: p.tool }) : say(WHAT["approval.bare"])) : WHAT[p.what] ? say(WHAT[p.what]) : p.what || ""
+      return say(TEXTS.waiting_for_person, { what })
+    }
+    case "waiting_for_subagents":
+      return say(TEXTS.waiting_for_subagents, { running: int(p.running) })
+    case "retrying": {
+      const left = Math.max(0, Math.trunc(Number(p.in) || 0) - Math.floor((now - Number(p.since || now)) / 1000))
+      return say(TEXTS.retrying, { in: String(left), reason: p.reason || "" })
+    }
+    case "stopped":
+      return has(p, "reason") ? say(TEXTS["stopped.reason"], { reason: STOPPED[p.reason] ? say(STOPPED[p.reason]) : p.reason }) : say(TEXTS.stopped)
+    case "failed":
+      return say(TEXTS.failed, { reason: p.reason || "" })
+    default:
+      return say(TEXTS[p.phase])
+  }
+}
+
+/**
+ * The line: `{{text}} · {{elapsed}}`, the text alone for stopped and failed (and where `bare`:
+ * a row with no room for the clock).
+ */
+export function progressLine(p, now = Date.now(), { bare = false } = {}) {
+  const text = progressText(p, now)
+  if (!text || bare || NO_ELAPSED.includes(p.phase) || !has(p, "since")) return text
+  // progress.line, `{{text}} · {{elapsed}}`: the same in every language.
+  return `${text} · ${elapsed(now - Number(p.since))}`
+}
+
+/** Whether the line changes by itself as time passes (it has a clock, or a countdown). */
+export function progressTicks(p) {
+  return isProgress(p) && !NO_ELAPSED.includes(p.phase) && has(p, "since")
+}

ui/js/scheduled.js

@@ -341,7 +341,7 @@ async function drawTask(id) {
         fact(t("Last run"), statusChip(task.last_run, task.running)),
         fact(t("Model"), modelWords(models, task.model)),
         fact(t("Reasoning effort"), h("span", { text: task.effort ? levelWord(task.effort) : t("The model's own") })),
-        fact(t("Agents"), h("span", { text: task.agents ? t("A team of up to %(n)s", { n: task.agents }) : t("No team") })),
+        fact(t("Subagents"), h("span", { text: task.agents ? t("Up to %(n)s at once", { n: task.agents }) : t("None") })),
         fact(t("Tell me"), h("span", { text: NOTIFY[task.notify]?.() || NOTIFY.matters() })),
         fact(t("Runs"), h("span", { text: MODE[task.mode]?.() || MODE.fresh() })),
       ),
@@ -582,15 +582,15 @@ async function drawForm(task) {
     const agents = h(
       "select",
       { id: "sched-agents", onchange: (e) => (draft.agents = Number(e.target.value) || null) },
-      h("option", { value: "", text: t("No team"), selected: !draft.agents }),
+      h("option", { value: "", text: t("None"), selected: !draft.agents }),
-      Array.from({ length: MAX_AGENTS - 1 }, (_, i) => i + 2).map((n) => h("option", { value: String(n), text: t("A team of up to %(n)s", { n }), selected: draft.agents === n })),
+      Array.from({ length: MAX_AGENTS - 1 }, (_, i) => i + 2).map((n) => h("option", { value: String(n), text: t("Up to %(n)s at once", { n }), selected: draft.agents === n })),
     )
     settings.replaceChildren(
       field(t("Tell me"), select("sched-notify", NOTIFY, "notify", () => (settings.querySelector("#sched-notify-hint").textContent = NOTIFY_HINT[draft.notify]())), NOTIFY_HINT[draft.notify]?.() || "", "sched-notify", "sched-notify-hint"),
       field(t("Runs"), select("sched-mode", MODE, "mode"), draft.mode === "continue" ? t("Each run sees the ones before it.") : t("Each run starts with nothing but the instruction."), "sched-mode"),
       field(t("Model"), modelSelect, t("A busy model is waited for, then the next free one is used."), "sched-model"),
       field(t("Reasoning effort"), effort, levels.length ? "" : t("Choose a model to choose its effort."), "sched-effort"),
-      field(t("Agents"), agents, t("A team works on it together, its own agent included."), "sched-agents"),
+      field(t("Subagents"), agents, t("How many work on it at once, the run's own agent among them, on the same model."), "sched-agents"),
     )
     settings.querySelector("#sched-mode").addEventListener("change", drawSettings)
   }

ui/js/terminal.js

@@ -0,0 +1,491 @@
+// The terminal panel: a shell on the device an agent chat works on, beside the chat -- the web
+// page's panel (the server's static/js/terminal.js), in this app. Two places it opens:
+//
+//   - an open agent chat: the chat's own shell (`{ chat }`), the one the web page shares;
+//   - a new agent chat, as soon as its device and folder are chosen (`{ device, directory }`):
+//     a shell there before the first message, which the chat that message makes takes over.
+//
+// The socket is Rust's (bridge.js `openTerminal`): the page never holds the token. xterm is the
+// server's vendored build (ui/vendor, scripts/fetch-vendor.sh), loaded the first time the panel
+// opens. Never beside the activity panel: the shell's layout has one panel column (as the web
+// page has), so opening one closes the other -- the shell stays open behind it.
+//
+// Three things about xterm the web panel learnt the hard way, kept here: it cannot fit while its
+// box is hidden (so every fit waits until the panel is shown), the panel changes size without a
+// window `resize` (so a ResizeObserver), and it reads no CSS variables (so the theme is read from
+// the computed tokens, and again when the theme changes).
+
+import { $, h, icon, ICONS } from "./dom.js"
+import { openTerminal, why } from "./bridge.js"
+
+const ANSI = ["black", "red", "green", "yellow", "blue", "magenta", "cyan", "white", "brightBlack", "brightRed", "brightGreen", "brightYellow", "brightBlue", "brightMagenta", "brightCyan", "brightWhite"]
+const MAX_MARKED = 200
+
+let ctx = { layout: () => {}, composer: () => {} }
+let target = null // what the panel is for: { key, chat } | { key, device, directory }, with `name`
+let term = null
+let fit = null
+let session = null // the open socket's handle, and which target it is for: { handle, key, draft }
+let observer = null
+let loading = null
+// A device's shell marks its own commands (OSC 133 A/B/C/D, OSC 7 for where it is) and says so
+// in its ready frame (`osc`): then a command and its output are read from the panel's own
+// buffer. Without the marks, "send" falls back to the screen as it is, and says so.
+let osc = false
+let pending = null
+let finished = []
+let visiting = -1
+
+const panel = () => $("terminal")
+
+function say(text, isError) {
+  const el = $("terminal-message")
+  el.textContent = text || ""
+  el.classList.toggle("is-error", !!isError)
+}
+
+/** Whether the panel is shown. */
+export function terminalOpen() {
+  return !panel().hidden
+}
+
+/** Whether there is a shell to open for what the chat page shows now. */
+export function terminalAvailable() {
+  return !!target
+}
+
+// --- xterm, loaded once --------------------------------------------------------------------------
+
+function load() {
+  if (loading) return loading
+  loading = new Promise((resolve, reject) => {
+    if (typeof window.Terminal === "function" && window.FitAddon) return resolve()
+    document.head.append(h("link", { rel: "stylesheet", href: "vendor/xterm.css" }))
+    const script = (src) =>
+      new Promise((ok, fail) => {
+        const s = h("script", { src })
+        s.addEventListener("load", ok)
+        s.addEventListener("error", () => fail(new Error(src)))
+        document.head.append(s)
+      })
+    script("vendor/xterm.js")
+      .then(() => script("vendor/xterm-addon-fit.js"))
+      .then(resolve, reject)
+  })
+  loading.catch(() => {
+    loading = null
+  })
+  return loading
+}
+
+function readTheme() {
+  const style = getComputedStyle(document.documentElement)
+  const token = (name, fallback) => (style.getPropertyValue(name) || "").trim() || fallback
+  const theme = {
+    background: token("--code-bg", "black"),
+    foreground: token("--ink", "white"),
+    cursor: token("--accent", "white"),
+    cursorAccent: token("--code-bg", "black"),
+    selectionBackground: token("--accent-soft", "gray"),
+  }
+  for (const name of ANSI) {
+    const value = token(`--ansi-${name.replace(/[A-Z]/g, (c) => `-${c.toLowerCase()}`)}`, "")
+    if (value) theme[name] = value
+  }
+  return theme
+}
+
+/** The theme changed: xterm's colours are read again. */
+export function terminalThemeChanged() {
+  if (term) term.options.theme = readTheme()
+}
+
+function build() {
+  if (term) return
+  const style = getComputedStyle(document.documentElement)
+  term = new window.Terminal({
+    allowProposedApi: true,
+    cursorBlink: true,
+    fontFamily: style.getPropertyValue("--font-mono").trim() || "monospace",
+    fontSize: parseFloat(style.getPropertyValue("--terminal-font-size")) || 13,
+    scrollback: 5000,
+    theme: readTheme(),
+  })
+  fit = new window.FitAddon.FitAddon()
+  term.loadAddon(fit)
+  term.open($("terminal-screen"))
+  term.parser.registerOscHandler(133, onCommandMark)
+  term.parser.registerOscHandler(7, onWhere)
+  term.onData((data) => session?.handle?.input(data))
+  // Ctrl+C is the shell's interrupt; copy and paste are the shifted pair, as in every terminal.
+  term.attachCustomKeyEventHandler((e) => {
+    if (!e.ctrlKey || !e.shiftKey || e.type !== "keydown") return true
+    if (osc && (e.key === "ArrowUp" || e.key === "ArrowDown")) {
+      visit(e.key === "ArrowUp" ? -1 : 1)
+      return false
+    }
+    const key = e.key.toLowerCase()
+    if (key === "c") {
+      const text = term.getSelection()
+      if (text) navigator.clipboard.writeText(text).catch(() => {})
+      return false
+    }
+    if (key === "v") {
+      navigator.clipboard
+        .readText()
+        .then((text) => session?.handle?.input(text))
+        .catch(() => {})
+      return false
+    }
+    return true
+  })
+  observer = new ResizeObserver(() => refit())
+  observer.observe($("terminal-screen"))
+}
+
+function refit() {
+  if (!term || !fit || !terminalOpen() || !$("terminal-screen").offsetWidth) return
+  try {
+    fit.fit()
+  } catch {
+    return
+  }
+  session?.handle?.resize(term.cols, term.rows)
+}
+
+// --- The socket ------------------------------------------------------------------------------------
+
+function decode(b64) {
+  const raw = atob(b64)
+  const bytes = new Uint8Array(raw.length)
+  for (let i = 0; i < raw.length; i++) bytes[i] = raw.charCodeAt(i)
+  return bytes
+}
+
+async function connect() {
+  if (!target || !term || session) return
+  const t0 = target
+  // The session exists before the socket answers, so frames that come first are this one's.
+  const mine = { handle: null, key: t0.key, draft: !t0.chat }
+  session = mine
+  say(t("Connecting…"))
+  let handle
+  try {
+    handle = await openTerminal(t0.chat ? { chat: t0.chat } : { device: t0.device, directory: t0.directory }, term.cols || 80, term.rows || 24, {
+      onOutput: (b64) => {
+        if (session === mine) term.write(decode(b64))
+      },
+      onControl: (frame) => {
+        if (session === mine) control(frame)
+      },
+      onEnd: (end) => {
+        if (session !== mine) return
+        session = null
+        ended(end)
+      },
+    })
+  } catch (err) {
+    if (session === mine) session = null
+    say(why(err), true)
+    return
+  }
+  mine.handle = handle
+  // Let go of meanwhile (another chat, the panel's target changed): closed at once.
+  if (session !== mine) handle.close()
+}
+
+function control(frame) {
+  if (frame.t === "ready") {
+    say(frame.shared ? t("Connected. This shell is also open in another tab, and they share a size.") : t("Connected."))
+    if (frame.dir) $("terminal-where").textContent = frame.dir
+    osc = !!frame.osc
+    showMarkControls()
+    showLast("")
+    refit()
+    if (terminalOpen()) term.focus()
+  } else if (frame.t === "behind") {
+    // This window stopped reading and was let go so the others kept up. The scrollback is the
+    // state: connecting again costs nothing.
+    say(frame.message || t("Reconnecting…"))
+    drop()
+    term.reset()
+    forgetMarks()
+    connect()
+  } else if (frame.t === "closed" || frame.t === "error") {
+    // No reconnect of its own: a new shell has lost the folder, the environment and the
+    // half-typed command, and quietly putting one in its place is worse than saying it went.
+    say(frame.message || t("This terminal closed."), frame.t === "error")
+    drop()
+  }
+}
+
+function ended(end) {
+  if (end.code === 4401 || end.status === 401) say(t("Sign in again: the server no longer accepts this app's sign-in."), true)
+  else if (end.status === 404) say(t("The terminal needs a newer LLeMbas server. Ask its administrator to update it."), true)
+  else if (end.error) say(end.error, true)
+  else say(t("Disconnected. Close and reopen to reconnect."))
+}
+
+/** Let go of the socket (the shell on the device lives on, as the server keeps it). */
+function drop() {
+  const s = session
+  session = null
+  s?.handle?.close()
+}
+
+// --- What the panel is for -----------------------------------------------------------------------
+
+/**
+ * Tell the panel what the chat page shows: `{ chat, name, directory }` for an open agent chat,
+ * `{ device, directory, name }` for a new one with its device and folder chosen, or null (no shell
+ * to open). A new chat's shell becomes its chat's when the first message has made the chat: the
+ * server took that shell over, so the socket stays as it is.
+ */
+export function aimTerminal(next) {
+  const key = next ? (next.chat ? `chat:${next.chat}` : `device:${next.device}:${next.directory}`) : ""
+  const before = target
+  target = next ? { ...next, key } : null
+  if (target) {
+    $("terminal-title").textContent = target.name || t("Terminal")
+    $("terminal-where").textContent = target.directory || ""
+    $("terminal-where").title = target.directory || ""
+  }
+  if (before?.key === key) return
+  // The draft's shell is the new chat's now: the same socket, said for the chat from here on.
+  const adopted = session?.draft && before && !before.chat && target?.chat && before.directory === target.directory && before.device === target.device
+  if (adopted) {
+    session.key = key
+    session.draft = false
+    return
+  }
+  if (session) {
+    drop()
+    if (term) term.reset()
+    forgetMarks()
+    say("")
+  }
+  if (!target) {
+    if (terminalOpen()) showTerminal(false)
+    return
+  }
+  if (terminalOpen()) connect()
+}
+
+// --- Showing it ------------------------------------------------------------------------------------
+
+/** Show or hide the panel. Hiding keeps the shell and its socket: reopening is instant. */
+export async function showTerminal(on) {
+  if (on && !target) return
+  panel().hidden = !on
+  ctx.layout(on)
+  markOpen()
+  if (!on) return
+  try {
+    await load()
+  } catch {
+    say(t("The terminal could not be loaded."), true)
+    return
+  }
+  if (!terminalOpen()) return
+  build()
+  // A frame later: the panel has only just been shown and has no width to fit to yet.
+  requestAnimationFrame(() => {
+    refit()
+    connect()
+    term.focus()
+  })
+}
+
+export function toggleTerminal() {
+  showTerminal(!terminalOpen())
+}
+
+function markOpen() {
+  for (const b of document.querySelectorAll("[data-terminal-toggle]")) b.setAttribute("aria-pressed", String(terminalOpen()))
+}
+
+/** The header's button for an open chat; a new chat has the chip under the composer instead. */
+export function terminalButton() {
+  return h(
+    "button",
+    {
+      type: "button",
+      class: "tool",
+      "aria-label": t("Terminal"),
+      title: t("Terminal"),
+      "aria-controls": "terminal",
+      "aria-pressed": String(terminalOpen()),
+      "data-terminal-toggle": "",
+      onclick: toggleTerminal,
+    },
+    icon(ICONS.terminal),
+  )
+}
+
+// --- A device shell's own marks ----------------------------------------------------------------------
+
+function onCommandMark(data) {
+  if (!osc || !term) return false
+  const [kind, code] = String(data).split(";")
+  if (kind === "A") pending = { prompt: term.registerMarker(0) }
+  else if (kind === "B") {
+    pending = pending || {}
+    pending.input = term.registerMarker(0)
+    pending.inputX = term.buffer.active.cursorX
+  } else if (kind === "C") {
+    pending = pending || {}
+    pending.output = term.registerMarker(0)
+    pending.command = commandText(pending)
+  } else if (kind === "D") {
+    if (pending?.output) {
+      pending.exit = parseInt(code, 10) || 0
+      pending.end = term.registerMarker(0)
+      commandDone(pending)
+    }
+    pending = null
+  }
+  return true
+}
+
+function onWhere(data) {
+  if (!osc) return false
+  let path = String(data).replace(/^file:\/\/[^/]*/, "")
+  try {
+    path = decodeURIComponent(path)
+  } catch {}
+  if (path) $("terminal-where").textContent = path
+  return true
+}
+
+function linesBetween(from, to, firstX) {
+  const buffer = term.buffer.active
+  const out = []
+  for (let y = from; y < to; y++) {
+    const line = buffer.getLine(y)
+    if (!line) continue
+    const text = line.translateToString(true, y === from ? firstX || 0 : 0)
+    if (line.isWrapped && out.length) out[out.length - 1] += text
+    else out.push(text)
+  }
+  while (out.length && !out[out.length - 1].trim()) out.pop()
+  return out.join("\n")
+}
+
+function commandText(mark) {
+  if (!mark.input || !mark.output || mark.input.line < 0) return ""
+  return linesBetween(mark.input.line, Math.max(mark.output.line, mark.input.line + 1), mark.inputX).trim()
+}
+
+function outputOf(mark) {
+  if (!mark.output || !mark.end || mark.output.line < 0 || mark.end.line < 0) return ""
+  return linesBetween(mark.output.line, mark.end.line, 0)
+}
+
+function commandDone(mark) {
+  finished.push(mark)
+  if (finished.length > MAX_MARKED) finished.shift()
+  visiting = -1
+  showLast((mark.command || "…") + (mark.exit ? ` · exit ${mark.exit}` : ""))
+}
+
+function blockOf(mark) {
+  const where = $("terminal-where").textContent
+  const head = `${where ? `${where} ` : ""}$ ${mark.command || ""}`
+  return "```\n" + head + "\n" + outputOf(mark) + "\n```\n" + (mark.exit ? `exit ${mark.exit}\n` : "")
+}
+
+function lastFinished() {
+  for (let i = finished.length - 1; i >= 0; i--) if (finished[i].output && finished[i].output.line >= 0) return finished[i]
+  return null
+}
+
+function visit(step) {
+  const live = finished.filter((m) => (m.prompt || m.input || m.output)?.line >= 0)
+  if (!live.length) return
+  visiting = visiting < 0 ? live.length : visiting
+  visiting = Math.max(0, Math.min(live.length - 1, visiting + step))
+  const m = live[visiting]
+  term.scrollToLine((m.prompt || m.input || m.output).line)
+}
+
+function forgetMarks() {
+  finished = []
+  pending = null
+  visiting = -1
+  osc = false
+  showMarkControls()
+  showLast("")
+}
+
+function showMarkControls() {
+  $("terminal-prev").hidden = !osc
+  $("terminal-next").hidden = !osc
+}
+
+function showLast(text) {
+  $("terminal-last").textContent = text || ""
+}
+
+// --- Handing it to the chat --------------------------------------------------------------------------
+
+/** What "send" and "copy" take: the selection; else the last command (with marks); else the screen. */
+function capture() {
+  if (!term) return ""
+  const chosen = term.getSelection()
+  if (chosen && chosen.trim()) return "```\n" + chosen + "\n```\n"
+  if (osc) {
+    const mark = lastFinished()
+    if (!mark) {
+      say(t("Nothing has been run in this shell yet."))
+      return ""
+    }
+    return blockOf(mark)
+  }
+  const buffer = term.buffer.active
+  const last = buffer.baseY + buffer.cursorY
+  const lines = []
+  for (let y = Math.max(0, last - 40); y <= last; y++) {
+    const line = buffer.getLine(y)
+    if (line) lines.push(line.translateToString(true))
+  }
+  const text = lines.join("\n").replace(/\n+$/, "")
+  if (!text.trim()) {
+    say(t("Nothing to send: select some output first."))
+    return ""
+  }
+  say(t("Copied the last of the screen, as it appeared. This shell does not mark where commands begin."))
+  return "```\n" + text + "\n```\n"
+}
+
+/** Into the message box, never sent: what a machine printed is for the person to read first. */
+function sendToChat() {
+  const text = capture()
+  if (!text) return
+  ctx.composer(text)
+  // Where the panel covers the chat (a narrow window), it steps aside to show the message box.
+  if (window.matchMedia("(max-width: 64rem)").matches) showTerminal(false)
+}
+
+function copyOut() {
+  const text = capture()
+  if (!text) return
+  navigator.clipboard.writeText(text).then(
+    () => say(t("Copied.")),
+    () => {},
+  )
+}
+
+export function initTerminal({ onLayout, composer }) {
+  ctx = { layout: onLayout || (() => {}), composer: composer || (() => {}) }
+  $("terminal-icon").replaceChildren(icon(ICONS.terminal))
+  $("terminal-prev").replaceChildren(icon(ICONS.chevronUp))
+  $("terminal-next").replaceChildren(icon(ICONS.chevronDown))
+  $("terminal-copy").replaceChildren(icon(ICONS.copy))
+  $("terminal-send").prepend(icon(ICONS.send))
+  $("terminal-close").replaceChildren(icon(ICONS.close))
+  $("terminal-prev").addEventListener("click", () => visit(-1))
+  $("terminal-next").addEventListener("click", () => visit(1))
+  $("terminal-copy").addEventListener("click", copyOut)
+  $("terminal-send").addEventListener("click", sendToChat)
+  $("terminal-close").addEventListener("click", () => showTerminal(false))
+}

ui/js/tree.js

@@ -0,0 +1,200 @@
+// The subagent tree (harness spec 3.4.0, subagents.md §12) as the activity panel draws it, the
+// server's own panel in this app (`chat/_activity_tasks.html` tree_row): the main agent at the
+// root, every subagent under its parent in the order added, how many slots are open, then the
+// ended ones folded apart, dimmed, with how each ended. A row is the tasks' grid -- chevron, who
+// it is and what it is doing, its context use in the time column, a stop square -- and opened,
+// its facts (state, model, mode and request options, messages, findings), its task, its phase
+// line, the last room message to or from it, and the buttons to read its conversation and its
+// findings. Its subagents nest under it.
+//
+// The data is the server's `GET /api/v2/chats/{chat}/tree` (and the stream's `tree` frame):
+//
+//   { min, max, size, open_slots,
+//     agents: [{ id, name, position, parent, state, model, mode, request, manager, task,
+//                context: { used, window }, progress, findings, messages }],
+//     ended:  [{ ..., ended }] }
+//
+// Everything in it but the numbers is a model's words: set as text, never as markup.
+
+import { h, icon, ICONS } from "./dom.js"
+import { progressLine, progressText } from "./progress.js"
+
+/** The states' words (loop.json team.read.states; the server's TREE_STATE_WORDS). */
+const STATES = {
+  working: "working",
+  waiting: "waiting for a slot",
+  finished: "finished",
+  idle: "idle",
+  paused: "paused: context full",
+  done: "done",
+  out_of_context: "out of context",
+  stopped: "stopped",
+}
+/** A subagent that may still be stopped (the server's TREE_LIVE). */
+export const TREE_LIVE = ["working", "waiting", "paused"]
+
+export function treeState(state) {
+  return STATES[state] ? t(STATES[state]) : String(state || "")
+}
+
+/** floor(used × 100 / window), or null when the window is not known. */
+export function contextPercent(c) {
+  const used = Number(c?.used)
+  const window = Number(c?.window)
+  if (!(window > 0) || !(used >= 0)) return null
+  return Math.min(100, Math.floor((used * 100) / window))
+}
+
+/** The agents in tree order (an agent, then those under it, depth first), each with its depth. */
+export function treeOrder(agents) {
+  const names = new Map()
+  for (const a of agents) names.set(a.id, a.id).set(a.name, a.id)
+  const parentOf = (a) => {
+    const p = a.parent ? names.get(a.parent) : null
+    return p && p !== a.id ? p : null
+  }
+  const under = new Map()
+  for (const a of agents) {
+    const p = parentOf(a)
+    if (!under.has(p)) under.set(p, [])
+    under.get(p).push(a)
+  }
+  const out = []
+  const seen = new Set()
+  const walk = (parent, depth) => {
+    for (const a of under.get(parent) || []) {
+      if (seen.has(a.id)) continue
+      seen.add(a.id)
+      out.push({ agent: a, depth })
+      walk(a.id, depth + 1)
+    }
+  }
+  walk(null, 0)
+  for (const a of agents) if (!seen.has(a.id)) out.push({ agent: a, depth: 0 })
+  return out
+}
+
+/** The room's messages to or from an agent (by name). */
+function messagesOf(room, a) {
+  return room.filter((m) => m.from === a.name || (m.to || []).includes(a.name) || (m.recipients || []).includes(a.name))
+}
+
+function firstLine(text) {
+  return String(text || "").split("\n").find((l) => l.trim()) || ""
+}
+
+/**
+ * The tree as rows. `act` holds what the rows can do: `open(a)` (its conversation, read-only),
+ * `findings(a)`, `stop(a)`; `isOpen(key)` / `toggled(key, open)` keep unfolded rows unfolded
+ * across redraws. `room` is the room as read so far; `device` says the subagents run on the
+ * computer the chat runs on (an agent chat).
+ */
+export function treePart(tree, { room = [], act = {}, now = Date.now(), device = false } = {}) {
+  const box = h("div", { class: "subagents" })
+  box.append(h("p", { class: "tasks__team", text: t("%(size)s of %(max)s", { size: tree.size ?? (tree.agents || []).length, max: tree.max ?? "?" }) }))
+  if (Number(tree.open_slots) > 0) box.append(h("p", { class: "tasks__goal subagents__slots", text: t("Open slots: %(n)s", { n: tree.open_slots }) }))
+  if (device) box.append(h("p", { class: "tasks__goal", text: t("They run on the computer this chat runs on.") }))
+  box.append(h("div", { class: "tasks__list subagents__list" }, treeOrder(tree.agents || []).map(({ agent, depth }) => row(agent, depth, { room, act, now, ended: false }))))
+  const ended = tree.ended || []
+  if (ended.length) {
+    const key = "fold:ended"
+    const fold = h(
+      "details",
+      { class: "tasks__more subagents__ended", open: !!act.isOpen?.(key) },
+      h("summary", null, h("span", { class: "tasks__chev", "aria-hidden": "true" }, icon(ICONS.chevron)), h("span", { text: `${t("Ended")} ${ended.length}` })),
+      h("div", { class: "tasks__list" }, ended.map((a) => row(a, 0, { room, act, now, ended: true }))),
+    )
+    fold.addEventListener("toggle", () => act.toggled?.(key, fold.open))
+    box.append(fold)
+  }
+  return box
+}
+
+function row(a, depth, { room, act, now, ended }) {
+  const key = `sub:${a.id}`
+  const main = !a.parent && (a.id === "lead" || a.name === "lead")
+  const pct = ended ? null : contextPercent(a.context)
+  const said = messagesOf(room, a)
+  const last = said[said.length - 1]
+  const mode = a.mode ? (a.request?.length ? `${a.mode} + ${a.request.join(", ")}` : a.mode) : ""
+  const live = !ended && TREE_LIVE.includes(a.state)
+  const phase = !ended && a.progress ? progressText(a.progress, now) : ""
+  const stop =
+    live && !main && act.stop
+      ? h(
+          "button",
+          {
+            type: "button",
+            class: "agent__stop",
+            "aria-label": t("Stop %(name)s", { name: a.name }),
+            title: t("Stop %(name)s", { name: a.name }),
+            onclick: (e) => {
+              e.preventDefault()
+              e.stopPropagation()
+              act.stop(a)
+            },
+          },
+          h("span", { class: "agent__square", "aria-hidden": "true" }),
+        )
+      : h("span", { class: "agent__stop is-none", "aria-hidden": "true" })
+  const el = h(
+    "details",
+    { class: `agent subagent is-${String(a.state || "").replace(/[^a-z_]/g, "")}${ended ? " is-ended" : ""}`, open: !!act.isOpen?.(key), dataset: { id: a.id } },
+    h(
+      "summary",
+      { class: "agent__row" },
+      h("span", { class: "agent__chev", "aria-hidden": "true" }, icon(ICONS.chevron)),
+      h(
+        "span",
+        { class: "agent__what" },
+        h(
+          "span",
+          { class: "subagent__who" },
+          h("bdi", { class: "agent__name", text: a.name || a.id }),
+          a.id !== a.name ? h("span", { class: "agent__id", text: a.id }) : null,
+        ),
+        main ? h("span", { class: "agent__role", text: t("the main agent") }) : a.position ? h("span", { class: "agent__role", text: a.position }) : null,
+        // The phase alone where a row has no room for its clock (progress.md §3); ended: how.
+        h("span", { class: "subagent__phase", dataset: { phase: phase ? a.progress.phase : "" }, text: phase || treeState(a.state) }),
+      ),
+      h("span", { class: "agent__time", title: t("Context used"), text: pct === null ? "" : `${pct}%` }),
+      stop,
+    ),
+    h(
+      "div",
+      { class: "agent__body" },
+      h(
+        "p",
+        { class: "agent__facts" },
+        h("span", { class: "agent__state", text: treeState(a.state) }),
+        a.model ? h("span", { class: "agent__model", text: a.model }) : null,
+        mode ? h("span", { text: mode }) : null,
+        h("span", { text: t("messages: %(n)s", { n: a.messages ?? said.length }) }),
+        a.findings ? h("span", { text: t("findings: %(n)s", { n: a.findings }) }) : null,
+      ),
+      a.task && !main ? h("p", { class: "agent__brief", text: a.task }) : null,
+      phase ? h("p", { class: "agent__text subagent__line", dataset: { progress: JSON.stringify(a.progress) }, text: progressLine(a.progress, now) }) : null,
+      last ? h("p", { class: "agent__text", text: `${last.from} → ${(last.to || []).join(", ")}: ${firstLine(last.text)}` }) : null,
+      main
+        ? null
+        : h(
+            "p",
+            { class: "subagent__buttons" },
+            act.open ? h("button", { type: "button", class: "btn btn--sm", text: t("Open its conversation"), onclick: () => act.open(a) }) : null,
+            act.findings ? h("button", { type: "button", class: "btn btn--sm", text: t("Findings"), onclick: () => act.findings(a) }) : null,
+          ),
+    ),
+  )
+  el.style.setProperty("--depth", String(Math.max(0, Math.min(6, depth))))
+  el.addEventListener("toggle", () => act.toggled?.(key, el.open))
+  return el
+}
+
+/** The tree's phase lines move on by themselves between reads (their clock). */
+export function tickTree(root, now = Date.now()) {
+  for (const el of root.querySelectorAll(".subagent__line[data-progress]")) {
+    try {
+      el.textContent = progressLine(JSON.parse(el.dataset.progress), now)
+    } catch {}
+  }
+}

ui/tokens.css

@@ -1,4 +1,4 @@
-/* LLeMbas v2.0.0-beta.6 (a71e9b4bdb8a) static/css/tokens.css, copied by scripts/fetch-tokens.sh: do not edit.
+/* LLeMbas v2.0.0-beta.8 (d06ca353b801) static/css/tokens.css, copied by scripts/fetch-tokens.sh: do not edit.
    The font-face rules are left out; the app bundles no font. */
 /*
   Design tokens.

ui/vendor/xterm-addon-fit.js

@@ -0,0 +1,2 @@
+!function(e,t){"object"==typeof exports&&"object"==typeof module?module.exports=t():"function"==typeof define&&define.amd?define([],t):"object"==typeof exports?exports.FitAddon=t():e.FitAddon=t()}(self,(()=>(()=>{"use strict";var e={};return(()=>{var t=e;Object.defineProperty(t,"__esModule",{value:!0}),t.FitAddon=void 0,t.FitAddon=class{activate(e){this._terminal=e}dispose(){}fit(){const e=this.proposeDimensions();if(!e||!this._terminal||isNaN(e.cols)||isNaN(e.rows))return;const t=this._terminal._core;this._terminal.rows===e.rows&&this._terminal.cols===e.cols||(t._renderService.clear(),this._terminal.resize(e.cols,e.rows))}proposeDimensions(){if(!this._terminal)return;if(!this._terminal.element||!this._terminal.element.parentElement)return;const e=this._terminal._core,t=e._renderService.dimensions;if(0===t.css.cell.width||0===t.css.cell.height)return;const r=0===this._terminal.options.scrollback?0:e.viewport.scrollBarWidth,i=window.getComputedStyle(this._terminal.element.parentElement),o=parseInt(i.getPropertyValue("height")),s=Math.max(0,parseInt(i.getPropertyValue("width"))),n=window.getComputedStyle(this._terminal.element),l=o-(parseInt(n.getPropertyValue("padding-top"))+parseInt(n.getPropertyValue("padding-bottom"))),a=s-(parseInt(n.getPropertyValue("padding-right"))+parseInt(n.getPropertyValue("padding-left")))-r;return{cols:Math.max(2,Math.floor(a/t.css.cell.width)),rows:Math.max(1,Math.floor(l/t.css.cell.height))}}}})(),e})()));
+//# sourceMappingURL=addon-fit.js.map

ui/vendor/xterm.css

@@ -0,0 +1,218 @@
+/**
+ * Copyright (c) 2014 The xterm.js authors. All rights reserved.
+ * Copyright (c) 2012-2013, Christopher Jeffrey (MIT License)
+ * https://github.com/chjj/term.js
+ * @license MIT
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to deal
+ * in the Software without restriction, including without limitation the rights
+ * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+ * THE SOFTWARE.
+ *
+ * Originally forked from (with the author's permission):
+ *   Fabrice Bellard's javascript vt100 for jslinux:
+ *   http://bellard.org/jslinux/
+ *   Copyright (c) 2011 Fabrice Bellard
+ *   The original design remains. The terminal itself
+ *   has been extended to include xterm CSI codes, among
+ *   other features.
+ */
+
+/**
+ *  Default styles for xterm.js
+ */
+
+.xterm {
+    cursor: text;
+    position: relative;
+    user-select: none;
+    -ms-user-select: none;
+    -webkit-user-select: none;
+}
+
+.xterm.focus,
+.xterm:focus {
+    outline: none;
+}
+
+.xterm .xterm-helpers {
+    position: absolute;
+    top: 0;
+    /**
+     * The z-index of the helpers must be higher than the canvases in order for
+     * IMEs to appear on top.
+     */
+    z-index: 5;
+}
+
+.xterm .xterm-helper-textarea {
+    padding: 0;
+    border: 0;
+    margin: 0;
+    /* Move textarea out of the screen to the far left, so that the cursor is not visible */
+    position: absolute;
+    opacity: 0;
+    left: -9999em;
+    top: 0;
+    width: 0;
+    height: 0;
+    z-index: -5;
+    /** Prevent wrapping so the IME appears against the textarea at the correct position */
+    white-space: nowrap;
+    overflow: hidden;
+    resize: none;
+}
+
+.xterm .composition-view {
+    /* TODO: Composition position got messed up somewhere */
+    background: #000;
+    color: #FFF;
+    display: none;
+    position: absolute;
+    white-space: nowrap;
+    z-index: 1;
+}
+
+.xterm .composition-view.active {
+    display: block;
+}
+
+.xterm .xterm-viewport {
+    /* On OS X this is required in order for the scroll bar to appear fully opaque */
+    background-color: #000;
+    overflow-y: scroll;
+    cursor: default;
+    position: absolute;
+    right: 0;
+    left: 0;
+    top: 0;
+    bottom: 0;
+}
+
+.xterm .xterm-screen {
+    position: relative;
+}
+
+.xterm .xterm-screen canvas {
+    position: absolute;
+    left: 0;
+    top: 0;
+}
+
+.xterm .xterm-scroll-area {
+    visibility: hidden;
+}
+
+.xterm-char-measure-element {
+    display: inline-block;
+    visibility: hidden;
+    position: absolute;
+    top: 0;
+    left: -9999em;
+    line-height: normal;
+}
+
+.xterm.enable-mouse-events {
+    /* When mouse events are enabled (eg. tmux), revert to the standard pointer cursor */
+    cursor: default;
+}
+
+.xterm.xterm-cursor-pointer,
+.xterm .xterm-cursor-pointer {
+    cursor: pointer;
+}
+
+.xterm.column-select.focus {
+    /* Column selection mode */
+    cursor: crosshair;
+}
+
+.xterm .xterm-accessibility:not(.debug),
+.xterm .xterm-message {
+    position: absolute;
+    left: 0;
+    top: 0;
+    bottom: 0;
+    right: 0;
+    z-index: 10;
+    color: transparent;
+    pointer-events: none;
+}
+
+.xterm .xterm-accessibility-tree:not(.debug) *::selection {
+  color: transparent;
+}
+
+.xterm .xterm-accessibility-tree {
+  user-select: text;
+  white-space: pre;
+}
+
+.xterm .live-region {
+    position: absolute;
+    left: -9999px;
+    width: 1px;
+    height: 1px;
+    overflow: hidden;
+}
+
+.xterm-dim {
+    /* Dim should not apply to background, so the opacity of the foreground color is applied
+     * explicitly in the generated class and reset to 1 here */
+    opacity: 1 !important;
+}
+
+.xterm-underline-1 { text-decoration: underline; }
+.xterm-underline-2 { text-decoration: double underline; }
+.xterm-underline-3 { text-decoration: wavy underline; }
+.xterm-underline-4 { text-decoration: dotted underline; }
+.xterm-underline-5 { text-decoration: dashed underline; }
+
+.xterm-overline {
+    text-decoration: overline;
+}
+
+.xterm-overline.xterm-underline-1 { text-decoration: overline underline; }
+.xterm-overline.xterm-underline-2 { text-decoration: overline double underline; }
+.xterm-overline.xterm-underline-3 { text-decoration: overline wavy underline; }
+.xterm-overline.xterm-underline-4 { text-decoration: overline dotted underline; }
+.xterm-overline.xterm-underline-5 { text-decoration: overline dashed underline; }
+
+.xterm-strikethrough {
+    text-decoration: line-through;
+}
+
+.xterm-screen .xterm-decoration-container .xterm-decoration {
+	z-index: 6;
+	position: absolute;
+}
+
+.xterm-screen .xterm-decoration-container .xterm-decoration.xterm-decoration-top-layer {
+	z-index: 7;
+}
+
+.xterm-decoration-overview-ruler {
+    z-index: 8;
+    position: absolute;
+    top: 0;
+    right: 0;
+    pointer-events: none;
+}
+
+.xterm-decoration-top {
+    z-index: 2;
+    position: relative;
+}

ui/vendor/xterm.js

@@ -0,0 +1,2 @@
+!function(e,t){if("object"==typeof exports&&"object"==typeof module)module.exports=t();else if("function"==typeof define&&define.amd)define([],t);else{var i=t();for(var s in i)("object"==typeof exports?exports:e)[s]=i[s]}}(globalThis,(()=>(()=>{"use strict";var e={4567:function(e,t,i){var s=this&&this.__decorate||function(e,t,i,s){var r,n=arguments.length,o=n<3?t:null===s?s=Object.getOwnPropertyDescriptor(t,i):s;if("object"==typeof Reflect&&"function"==typeof Reflect.decorate)o=Reflect.decorate(e,t,i,s);else for(var a=e.length-1;a>=0;a--)(r=e[a])&&(o=(n<3?r(o):n>3?r(t,i,o):r(t,i))||o);return n>3&&o&&Object.defineProperty(t,i,o),o},r=this&&this.__param||function(e,t){return function(i,s){t(i,s,e)}};Object.defineProperty(t,"__esModule",{value:!0}),t.AccessibilityManager=void 0;const n=i(9042),o=i(9924),a=i(844),h=i(4725),c=i(2585),l=i(3656);let d=t.AccessibilityManager=class extends a.Disposable{constructor(e,t,i,s){super(),this._terminal=e,this._coreBrowserService=i,this._renderService=s,this._rowColumns=new WeakMap,this._liveRegionLineCount=0,this._charsToConsume=[],this._charsToAnnounce="",this._accessibilityContainer=this._coreBrowserService.mainDocument.createElement("div"),this._accessibilityContainer.classList.add("xterm-accessibility"),this._rowContainer=this._coreBrowserService.mainDocument.createElement("div"),this._rowContainer.setAttribute("role","list"),this._rowContainer.classList.add("xterm-accessibility-tree"),this._rowElements=[];for(let e=0;e<this._terminal.rows;e++)this._rowElements[e]=this._createAccessibilityTreeNode(),this._rowContainer.appendChild(this._rowElements[e]);if(this._topBoundaryFocusListener=e=>this._handleBoundaryFocus(e,0),this._bottomBoundaryFocusListener=e=>this._handleBoundaryFocus(e,1),this._rowElements[0].addEventListener("focus",this._topBoundaryFocusListener),this._rowElements[this._rowElements.length-1].addEventListener("focus",this._bottomBoundaryFocusListener),this._refreshRowsDimensions(),this._accessibilityContainer.appendChild(this._rowContainer),this._liveRegion=this._coreBrowserService.mainDocument.createElement("div"),this._liveRegion.classList.add("live-region"),this._liveRegion.setAttribute("aria-live","assertive"),this._accessibilityContainer.appendChild(this._liveRegion),this._liveRegionDebouncer=this.register(new o.TimeBasedDebouncer(this._renderRows.bind(this))),!this._terminal.element)throw new Error("Cannot enable accessibility before Terminal.open");this._terminal.element.insertAdjacentElement("afterbegin",this._accessibilityContainer),this.register(this._terminal.onResize((e=>this._handleResize(e.rows)))),this.register(this._terminal.onRender((e=>this._refreshRows(e.start,e.end)))),this.register(this._terminal.onScroll((()=>this._refreshRows()))),this.register(this._terminal.onA11yChar((e=>this._handleChar(e)))),this.register(this._terminal.onLineFeed((()=>this._handleChar("\n")))),this.register(this._terminal.onA11yTab((e=>this._handleTab(e)))),this.register(this._terminal.onKey((e=>this._handleKey(e.key)))),this.register(this._terminal.onBlur((()=>this._clearLiveRegion()))),this.register(this._renderService.onDimensionsChange((()=>this._refreshRowsDimensions()))),this.register((0,l.addDisposableDomListener)(document,"selectionchange",(()=>this._handleSelectionChange()))),this.register(this._coreBrowserService.onDprChange((()=>this._refreshRowsDimensions()))),this._refreshRows(),this.register((0,a.toDisposable)((()=>{this._accessibilityContainer.remove(),this._rowElements.length=0})))}_handleTab(e){for(let t=0;t<e;t++)this._handleChar(" ")}_handleChar(e){this._liveRegionLineCount<21&&(this._charsToConsume.length>0?this._charsToConsume.shift()!==e&&(this._charsToAnnounce+=e):this._charsToAnnounce+=e,"\n"===e&&(this._liveRegionLineCount++,21===this._liveRegionLineCount&&(this._liveRegion.textContent+=n.tooMuchOutput)))}_clearLiveRegion(){this._liveRegion.textContent="",this._liveRegionLineCount=0}_handleKey(e){this._clearLiveRegion(),/\p{Control}/u.test(e)||this._charsToConsume.push(e)}_refreshRows(e,t){this._liveRegionDebouncer.refresh(e,t,this._terminal.rows)}_renderRows(e,t){const i=this._terminal.buffer,s=i.lines.length.toString();for(let r=e;r<=t;r++){const e=i.lines.get(i.ydisp+r),t=[],n=e?.translateToString(!0,void 0,void 0,t)||"",o=(i.ydisp+r+1).toString(),a=this._rowElements[r];a&&(0===n.length?(a.innerText=" ",this._rowColumns.set(a,[0,1])):(a.textContent=n,this._rowColumns.set(a,t)),a.setAttribute("aria-posinset",o),a.setAttribute("aria-setsize",s))}this._announceCharacters()}_announceCharacters(){0!==this._charsToAnnounce.length&&(this._liveRegion.textContent+=this._charsToAnnounce,this._charsToAnnounce="")}_handleBoundaryFocus(e,t){const i=e.target,s=this._rowElements[0===t?1:this._rowElements.length-2];if(i.getAttribute("aria-posinset")===(0===t?"1":`${this._terminal.buffer.lines.length}`))return;if(e.relatedTarget!==s)return;let r,n;if(0===t?(r=i,n=this._rowElements.pop(),this._rowContainer.removeChild(n)):(r=this._rowElements.shiftLine truncated
+//# sourceMappingURL=xterm.js.map

vendor.lock.json

@@ -1,6 +1,6 @@
 {
-  "tag": "v2.0.0-beta.6",
+  "tag": "v2.0.0-beta.8",
-  "commit": "a71e9b4bdb8acb6e9949e978253a72c347cc67f3",
+  "commit": "d06ca353b801916302f0ce0cc9ae3e5244ff184f",
   "files": {
     "cm-autocomplete.mjs": {"server": "26fe431a5f54e29952e094c93151f11473a80677bfb339711ecca08195a89df4", "copy": "fcb7fcbe4e2b072d031520ec898c81ac3f2f05fb4126e03003c9b5e60ffaa8de"},
     "cm-codemirror.mjs": {"server": "dea86a106e13c9316dfedfd29209780b853bdb2b477edf3f7c90f497b592a223", "copy": "0b80ae1289a1dc3e6718727e78bc757caad657e907d0351597867d4f90c36bd2"},
@@ -19,6 +19,9 @@
     "lezer-common.mjs": {"server": "b63d03ca460cc7a6ce09cbf0de24943fec22bcbe2359dc35978a57b875b2450f", "copy": "b63d03ca460cc7a6ce09cbf0de24943fec22bcbe2359dc35978a57b875b2450f"},
     "lezer-highlight.mjs": {"server": "b74148d673103ae08330a045a464acddf4ee8b563b213ea26e50f2be01a6d19d", "copy": "df21a1741ae80f2ad2b6d25860cad9ff3cda246a4fba171535d9dc92df150c1a"},
     "lezer-lr.mjs": {"server": "4c2dcc61b5c3aedb091dfd60ce8abaeb94704b58fa9036547b0b8d2c75f701e4", "copy": "f41bc21dc3fd874132e1d9862aae57946ae2604d91330303924c03aecfd3bc21"},
+    "xterm.js": {"server": "1f991ac3b4b283ebf96e60ae23a00a52765dd3a2e46fa6fdda9f1aab032f7495", "copy": "1f991ac3b4b283ebf96e60ae23a00a52765dd3a2e46fa6fdda9f1aab032f7495"},
+    "xterm.css": {"server": "ba8e6985669488981ccf40c0cefe3aba80722cb6c92de7ad628b0bd717faf2b6", "copy": "ba8e6985669488981ccf40c0cefe3aba80722cb6c92de7ad628b0bd717faf2b6"},
+    "xterm-addon-fit.js": {"server": "bdaefa370b1bfc42ee88d46fe6072400902a4d4b2d45cd93438dda9b23c97089", "copy": "bdaefa370b1bfc42ee88d46fe6072400902a4d4b2d45cd93438dda9b23c97089"},
     "node-process.mjs": {"server": "0c7460cbd81d8427a882867726b6db43beff3399f045ad942f4a65fef4705981", "copy": "0c7460cbd81d8427a882867726b6db43beff3399f045ad942f4a65fef4705981"}
   }
 }