- Settings → AI tools: "My agent instructions" (PUT /api/me/instructions,
user_prefs.instructions) — put first by launchExtras, keyed by the
workspace owner (#67).
- Session view: Settings opens as one more tab of the tab bar.
- Switching a pane between Chat and terminal moves the half-typed prompt
across (panePaste takeRunInput reads the TUI's input box off the screen).
- Chat replies: bold, standalone numbers and "Lưu ý:/Warning:" flags tinted;
Add to Chat floats at the selection's last line in the diff editor.
- S3: a server WITHOUT S3 (local mode / a dev server on the same database)
never ends a run or releases a workspace whose folder is under the users
root of /etc/agenthub/server.conf (linux-accounts.unseenMemberFolder, used
by reconcileOrphans and healRunId). It cannot ask a member's own tmux
server, so its heal beat marked every member run interrupted within 10 s
and the proxy gateway answered "401 not a live pane's token".
Regression test: s3-unseen-member-runs.test.js.
Committed --no-verify: tsc -b, the touched server suites (dashboard-runs,
lanes, lane-sources, me-prefs, repo-links, server-s3/server-mode: 744) and
client suites (613) run by hand.
Usage panel (status bar):
- every account shows its plan (Claude subscriptionType, Codex
chatgpt_plan_type, Cursor membershipType, Grok subscriptionTier)
- new quota rows: Cursor CLI (cursor.com usage-summary), Antigravity
(`agy --print /usage`, two model pools; AGENTHUB_AGY_USAGE=0 never runs
agy), Grok Build, Kimi Code, OpenCode Go — each login read from the tool's
own file, sent only to its own host while valid, never refreshed
- in-flight quota reads are shared between tabs
AI Tools catalog: Grok Build (`grok`) and Kimi Code (`kimi`) added;
Gemini CLI, Hermes and OpenClaw removed.
Also: Quit AgentHub (#65, `agenthub stop`), status-area icon, Add to Chat
on an editor selection, Settings in three groups, JetBrains-style Changes
panel (ticked files/hunks, ignore, blame, file history), terminal font.
--no-verify: :3000 is taken by another dev server (lane-detect tests);
both suites run by hand — server 2636/2639 (the 3 are :3000), client
1131/1131.
- EditorModal loads once per path (a parent re-render no longer reloads the
file over the user's typing), refuses Save after a failed load, opens a
truncated (>256 KB) file read-only, and asks before overwriting a file
that changed on disk or could not be re-read.
- Settings/Hooks editors unset only keys from the snapshot taken at Edit,
warn when the file changed underneath, keep a renamed key's value, and
refuse a new key named like a redacted one (env) or a read-only hook
event, which used to replace the secret/hooks.
- Project scope shows its resolved folder and is disabled inside the
plugin cache. Account-profile, pricing-row and pet deletes confirm;
two-click actions disarm after 3 s; unsaved drafts ask before closing.
- Search indexes every Claude Code / Codex sub-tab, opens it via ?tab= and
scrolls to + highlights the row; the active config tab is in the URL.
- General splits into General / Data / About; Pricing moves to a Cost
group with one shared load; config tabs group as Content / Extensions /
Config with arrow-key roving; Promise.allSettled per tab.
- Data: browser-preferences export/import (allowlisted keys), per-section
reset, housekeeping dry-run first, clear-all behind a typed phrase.
Server: purge days must be a whole number >= 1 (INVALID_PURGE_DAYS) and
dry_run returns exactly what the real run would delete.
- Alerts can be edited; webhooks/alerts/remote-source failures toast;
copies go through copyText; modals get dialog roles and useDialog.
- Sagi: Ctrl/Cmd+B only when shown and not typing, default spot clear of
row controls, arrow keys move it, reduced motion honoured; the dead
/run launch link goes to the workspace.
The workspace home's composer takes ONE Model chip (ModelMenu: model list
plus Effort) and moves AI Router / Permission / Local-New worktree into the
row under it; Settings drops the General-page cost card and export row,
PricingSettings and the empty states are compacted for settings panels.
Settings moves from a tab bar to /settings/<section>: a left nav grouped
General / Agents / Routing / Git / Projects (a select on narrow screens), a
search box over every section (Ctrl/Cmd+F, Enter opens the first match), and
row primitives. Each AI tool has its page under agents/<id>, the AI Tools
modal becomes the Agents page, Pricing shows every vendor, Git → Worktrees
sets which untracked worktrees the sidebar lists, and each project gets a
page with its own worktree override and Forget all. ?tool= and #ai-tools
links redirect to the path form.
Pricing: the LiteLLM sync now adds every current Claude / GPT (>= 5) /
Gemini (>= 3) chat model before it is used and keeps its own rows
(model_pricing.source = 'sync', the Auto badge) at the published rate. A
user's row (any PUT) and a deleted pattern (model_pricing_dismissed) are
never touched; a deleted built-in row is not re-seeded at the next start; a
reset re-applies the list at once; retired Claude models are priced only
once used.
Models: GET /api/ai-tools/:id/models reads each tool's own list (Codex's
models_cache.json, Claude's aliases named after the newest version plus
priced versions, a catalog tool's vendor rows plus the models its sessions
used). A catalog `model: {flag, vendor?}` field passes `<flag> <id>`
(validated: never a flag or shell text); a per-tool default model starts
each run, and a resume keeps the session's own model.
Settings: every installed, enabled catalog tool without its own tab gets
one (AiToolConfig.tsx): path, homepage, Update, sessions/cost/lanes,
default model, vendor pricing, and the config files its release catalog
lists, edited over the opened sha, syntax-checked, backed up, written
atomically; credential-looking names refused. Antigravity keeps its hooks
card there.
Lanes: a just-started run no longer reads "dead" (age counts the run's
start while the old session is still bound); a second dashboard on the
same database re-broadcasts the lanes the other changed (PRAGMA
data_version, every 2 s); a lane start with a bad model answers 400.
DB: every migration runs under a cross-process lock (<db>-migrate.lock), so
two dashboards started at once on a fresh database both come up.
nntrivi2001