Skip to content

feat(threads): first-class standing bot threads - #110

Merged
nullStack65 merged 6 commits into
release/native-0.1.10from
feat/standing-bot-threads
Nov 19, 2026
Merged

nullStack65 merged 6 commits into
release/native-0.1.10from
feat/standing-bot-threads

Conversation

@nullStack65

@nullStack65 nullStack65 commented Nov 19, 2026 •

Copy link
Copy Markdown
Owner

Standing bot threads (Lead, Product Designer, CEO, CTO, CFO) looked the same as pinned watch threads: the same project, branch, status and time on every row. This PR makes "standing" a real thread property with a bot persona and gives bots their own compact look.

What changed

  1. Data. A thread carries standing: { name, role, avatarId, accent } | null, set by the new thread.standing.set command and the thread.standing-updated event. It persists, syncs to every client and is independent of pinning. Setting it does not bump updatedAt. Old clients ignore the event. The threadStanding environment capability gates the UI.
  2. Sidebar. A Bots group above Pinned holds one 32px row per bot: avatar, name, latest status or role line, and a dot badge. Branch, worktree, PR, model and time stay off these rows. Bots sort by name so their places never move.
  3. Avatar states. Idle is still. Running blinks the eyes on a 4s stepped duty cycle (step-end, no continuous repaint), and prefers-reduced-motion turns that off. Needs-you adds an accent ring plus an accent badge. Failed adds a destructive ring.
  4. Bot thread view. The header shows the avatar, name and role in place of the project/thread breadcrumb. The details card (host, branch, worktree, PR) folds the first time a session opens a bot, and the existing panel toggle reopens it. The composer's branch strip hides while it is folded.
  5. Setting it. The thread menu has Make standing bot…, Edit bot persona… and Make ordinary thread. A dialog edits the name, role, one of 10 avatars and one of 8 accents. t3_thread_update takes action: "set_standing" with standing (a persona, or null to clear). t3_thread_launch takes an optional standing, applied after launch with a derived command id so a retry stays idempotent.

Screenshots

These use synthetic data on an isolated dev server on R720, in headless Chromium.

Light Dark
Before: the bots are pinned and look like the watch thread before light before dark
After: the Bots group above an ordinary pinned watch thread after light after dark
Running: "Ops" is mid-turn and its eyes blink running light running dark
Bot thread: slim header, details card folded thread light thread dark

Browser check:

  1. Four rows were made bots through the real menu and dialog.
  2. The running state produced 1 bot-blink animation, and 0 under reducedMotion: reduce.
  3. The needs-you and failed rings were not seen live, because the replay provider auto-approves in full access. BotAvatar.test.tsx covers them instead.

Prior art

I looked at these before drawing the avatars:

  1. DiceBear (MIT code; each style has its own license, some CC BY 4.0 with attribution). It is a good fit for generated avatars, but it is a dependency plus per-style license tracking for 10 small icons.
  2. Boring Avatars (MIT). Its abstract shapes are clean, but they don't read as "bot".
  3. Multiavatar. Its license is custom and non-standard for redistribution, and it is too busy at 20px.
  4. Lottie / Rive. Animated, but it needs a runtime, and continuous playback conflicts with this repo's no-repaint rule.
  5. Original SVG (chosen). Ten 24×24 line drawings in currentColor, so the accent and theme set the color. The eyes sit in one <g> so a CSS step animation can blink them. There is no dependency and no third-party art. Nothing is copied from Grok, xAI or fan packs.

Third-party notices: third-party-licenses.config.json has a Standing bot avatars entry stating that the art is original and ships under this repo's MIT license.

Tests

  1. threadStanding.test.ts (contracts) covers resolving and trimming personas.
  2. runtimeLayer.test.ts covers set, clear, the projection and the unchanged updatedAt.
  3. tools.test.ts covers the set_standing validation.
  4. botPresence.test.ts, BotAvatar.test.tsx and threadActionMenu.logic.test.ts cover the web logic.
  5. Typecheck is clean for contracts, client-runtime, server, web and mobile.

Base note: this targets release/native-0.1.10, not main. The fleet builds from the release branch, and the MCP thread tools only exist there.

Deploy: fleet

Model: claude-opus-5-5
Task: feature

🤖 Generated with Claude Code

Adds an optional `standing` persona (name, role, avatar id, accent) to V2
threads and shells, a `thread.standing.set` command, the
`thread.standing-updated` event, and the `threadStanding` capability.
Standing is independent of pinning.

Model: claude-opus-5-5
Task: feature
Agent: claude-code

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@nullStack65

Copy link
Copy Markdown
Owner Author

START: first-class standing bot threads

Issues are disabled on nullStack65/t3code, so this PR is where the START, checkpoint and RESULT comments go.

Scope:

  1. A standing persona on threads (name, role, avatar, accent). It is independent of pinning.
  2. A thread.standing.set command and the threadStanding capability.
  3. A compact Bots group above Pinned in the web sidebar, with avatar states: idle, running, needs-you and failed.
  4. A slim header for bot threads, with branch, worktree and PR detail folded into a collapsible details section.
  5. A "Make standing bot" / "Make ordinary thread" menu action with persona editing.
  6. t3_thread_launch and t3_thread_update accept standing and persona.
  7. Original in-repo SVG bot avatars, plus the prior-art survey.
  8. Rollout through the existing fleet update path.
  9. Mark these threads on the laptop (environment 2f360b2e…) as bots:
    • Product Designer (mcp:launch:859a4c02…)
    • CEO (mcp:launch:f87490e4…)
    • CTO and CFO, if they exist by rollout. Search for those titles from launcher 91722712…; nothing else on those threads changes.

Base: release/native-0.1.10 @ 440e116
Head: 5c89445 (contracts only so far)

Adds the thread.standing.set command and thread.standing-updated event.
Standing is identity, not activity, so it never bumps updatedAt.

Model: claude-opus-5-5
Task: standing bot threads
Agent: Claude Code

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@nullStack65

Copy link
Copy Markdown
Owner Author

CHECKPOINT head cf60874: server wiring for thread.standing.set / thread.standing-updated (Orchestrator, ProjectionStore, relay, client projection). vp test run src/orchestration-v2/runtimeLayer.test.ts -t lifecycle: 118 passed. Server typecheck exit 0. Next: client shell + MCP tools + web UI.

nullStack65 and others added 3 commits November 19, 2026 19:01
Adds the set_standing update action, a standing launch parameter, the
client shell field and the web setThreadStanding action gated on the
threadStanding capability.

Model: claude-opus-5-5
Task: standing bot threads
Agent: Claude Code

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Standing bots list under a Bots group above Pinned with one-line rows
(avatar, name, status or role, badge). Ten original SVG avatars blink
on a stepped duty cycle while running and ring for needs-you or failed.
Bot threads get a slim header with branch, worktree and PR behind a
Details toggle. The thread menu can make, edit or revert a bot.

Model: claude-opus-5-5
Task: standing bot threads
Agent: Claude Code

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The branch, worktree and PR live in the thread details card, so a bot
folds that card the first time a session shows it and the existing
panel toggle reopens it. Adds avatar render tests.

Model: claude-opus-5-5
Task: standing bot threads
Agent: Claude Code

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@nullStack65

Copy link
Copy Markdown
Owner Author

CHECKPOINT head 4dff27f: web UI complete and browser-checked.

  1. Added: the Bots sidebar group, 10 original avatars, the bot persona dialog, the thread-menu actions, the slim bot header (details card folds on first view), the notices entry, and logic and render tests.
  2. Checks: web typecheck and lint are clean, and the web unit tests pass (botPresence, BotAvatar, threadActionMenu). Mobile and client-runtime typecheck clean.
  3. Browser: headless Chromium against an isolated dev server on R720 with synthetic data. Four rows were made bots through the real menu and dialog. Captured light and dark shots, and the running blink measured 1 animation, 0 under reduced motion. Needs-you was not seen live, because the replay auto-approves.
  4. Screenshots: orphan branch pr-assets/standing-bots at e1eb50d, embedded in the PR body.
  5. Next: one independent review on claude-fable-5-2.

… review fixes

Addresses the independent claude-fable-5-2 review round:
- standing bots are never auto-settled and never pruned as older settled shells
- sidebar bot split is a tested pure helper; bots are excluded from drag keys
- t3_thread_launch reports standingApplied instead of failing silently
- bot details fold state persists per thread; header remeasures on standing change
- persona dialog swatch/avatar buttons get distinct accessible names

Model: claude-opus-5-5
Task: standing bot threads (nullStack65/t3code PR #110)
Agent: Claude Code
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@nullStack65

Copy link
Copy Markdown
Owner Author

CHECKPOINT — review fixes, head 1815eae8d2080be5f13598af7b573cb17eabbda0

One independent review round (claude-fable-5-2). Every finding is fixed in this commit:

  1. Settled bots could disappear: auto-settlement now skips standing threads (ThreadSettlementService, ProjectionStore SQL), and isOlderSettledThreadShell never prunes a bot. Tests added.
  2. Sidebar partition coverage: the bot split is now a pure helper, splitStandingThreads. Its unit test covers pinned, settled and snoozed bots.
  3. A launch that silently half-failed: t3_thread_launch now returns standingApplied and logs mcp.thread_launch.standing_failed.
  4. Drag keys: bots are excluded from reorderable keys. Bots that are delegated children stay nested under their parent.
  5. A stale header measure: standing was added to the breadcrumb effect's dependencies.
  6. Fold state was lost on reload: it now persists in localStorage, capped at 500 entries.
  7. A11y nit: avatar and accent buttons now have distinct names.

Results: web unit tests 105/105, server tests 27/27 (settlement, orchestrator tools, project handlers), contracts tests 3/3. Web typecheck and lint are clean.

@nullStack65
nullStack65 marked this pull request as ready for review November 19, 2026 19:30
@nullStack65
nullStack65 merged commit b424e48 into release/native-0.1.10 Nov 19, 2026
3 of 5 checks passed
@nullStack65

Copy link
Copy Markdown
Owner Author

CHECKPOINT: merged. Head 1815eae8d2080be5f13598af7b573cb17eabbda0, squash merge b424e48ea1b7dd76fe05bdafaeded7d40622a0f8. I used --match-head-commit without --admin.

These checks passed at the head: Voice + founder inbox (Linux ARC) took 7m12s, and the Deploy line and CodeRabbit checks also passed (CodeRabbit was skipped for this base branch).

Rollout plan:

  1. The R720 and ASUS servers go through the fleet reconciler once the push CI qualifies the release head.
  2. The laptop desktop needs a pin bump in closura-agent-config. The AppImage is building on primary-r720 from b424e48e now. Applying the pin stages the build only, and it takes effect when T3 is next reopened.

@nullStack65

Copy link
Copy Markdown
Owner Author

CHECKPOINT: rollout status (2026-11-19, about 22:05 UTC)

  1. Servers, R720 and ASUS. Both still run 1a634c53. The reconciler has seen release head 7af284f1, which contains this merge (b424e48e), and is waiting for its ARC CI run to qualify before it builds and deploys. No manual action is taken; the reconciler owns this.
  2. Laptop desktop. Pin-bump PR nullStack65/closura-agent-config#933 (head 85095f4c) stages build b424e48e. All 11 checks pass and an independent review found no blockers. Because of the merge hold (Project folders with emoji may cause startup/work log errors and broken file links pingdotgg/t3code#906), it is labeled merge-batch and not merged. After it merges, run the t3-release recipe apply on the laptop. That only stages the build; it loads when T3 is next quit and reopened.
  3. Bot marking. CEO, CTO, CFO and Product Designer are not marked yet. All four are on the laptop and were checked against launcher 91722712. Marking needs the laptop's server to run b424e48e or newer.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant