Repository navigation
Friend Bot row: full-width hover; fallback avatar never matches own (part of #384) - #402
Conversation
…own (ThinkFlowLab#384) The settings gear on a wrapped Bot row is absolutely positioned so hover and selected fill the same full-width band as a plain row (including at 150% zoom). A friend's Bot with no published look steps its seeded shape when it would match one of your own Bots exactly; a stored image/emoji/shape is left alone. Docs-not-needed: subsystem note updated in docs/subsystems/core-client.md.
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 8d6825275d
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| const settled = useSettled(state) | ||
| if (bot === undefined) return h(BotMark, { size }) | ||
| // A friend's Bot never shares the exact fallback avatar of one of my own Bots (#384). | ||
| const look = bot.remote ? lookAvoidingLocals(bot, locals) : undefined |
There was a problem hiding this comment.
When a remote fallback collides, this only works where a caller supplies locals; an omitted prop becomes []. CommandPalette (core/src/client/palette.js:54) and the details drawer (core/src/client/details-drawer.js:306) call BotAvatar without it, while remote chat messages use BotMark directly (core/src/client/cells.js:273). Consequently the same friend is distinct in the sidebar/header but reverts to the colliding seed on these surfaces, so pass the own roster through every renderer (including direct marks) or centralize avoidance.
Useful? React with 👍 / 👎.
| export function lookKey(look) { | ||
| if (look?.image) return `image:${String(look.image).slice(0, 96)}` | ||
| if (look?.emoji) return `emoji:${look.emoji}` | ||
| return `shape:${shapeId(look?.shape)}:${look?.color ?? ''}` |
There was a problem hiding this comment.
With a partially published remote look (color present, no avatar shape), lookKey compares raw identifiers rather than rendered colors. In light mode an own Bot colored blue renders #2A92FE (core/src/client/inks.js:8), while Bot Social normalizes that published color to #2a92fe (plugins/social/src/look.js:35-36); identical shapes therefore produce different keys and the helper returns the visually identical fallback. Normalize equivalent named/hex and hex-case values in the key.
Useful? React with 👍 / 👎.
| const candidate = { ...look, shape: shapes[index] } | ||
| if (!taken.has(lookKey(candidate))) return candidate | ||
| } | ||
| return look |
There was a problem hiding this comment.
For a team whose own Bots occupy all 19 non-whale picker shapes in the fallback's color, this loop necessarily exhausts: line 120 established the original key is taken, yet this post-loop return gives that same colliding look back. Nested tree rosters can exceed this size, so this is a valid configured team and the advertised “never matches” guarantee fails; vary another dimension or otherwise return an untaken key after shape exhaustion.
AGENTS.md reference: AGENTS.md:L69-L69
Useful? React with 👍 / 👎.
…endered colors, step color after shapes (ThinkFlowLab#384)
|
Addressed the three Codex P2 findings at
Still needs an independent reviewed-tip PASS (T1) before the merge queue. |
…ew (ThinkFlowLab#426) Owner decision 2026-10-19 15:31 (Asia/Shanghai): stop using Codex for PR review. The owner's own reviewer Bot, acting as `Birfy`, reviews and PASSes (self-PASS already allowed by ThinkFlowLab#415, `selfReview: ["Birfy"]`). How Codex was triggered: not by anything in this repo. No workflow, script, routine, label or `@codex` comment requests it (rg of `.github/`, skills, scripts, docs; `gh api repos/ThinkFlowLab/FlyBot/hooks` is empty). The reviews on ThinkFlowLab#402/ThinkFlowLab#414/ThinkFlowLab#415/ThinkFlowLab#416 come from the `chatgpt-codex-connector` GitHub App with trigger "PR opened", i.e. Codex's automatic code review enabled for this repo in ChatGPT Codex settings. That switch is outside the repo and is turned off there by the owner. This PR (wording only, no gate change): - AGENTS.md, CONTRIBUTING.md, docs/autonomous-dev.md, LESSONS.md: the reviewer is the owner's Bot (self-review); nothing requests or waits for Codex. - review-pr skill: drop the OpenCode↔Codex agreement requirement and Verifier line; reviewer's own evidence at the tip decides. - review-gate.test.mjs: neutral fixture login `reviewer-agent` instead of `codex-agent`, drop the unused agreement line (the gate never parsed it). Unchanged: `.github/agent/review-policy.json` (reviewers `hsliuustc0106`, `Birfy`; selfReview `Birfy`; Codex was not listed), CI, risk tiers, T2/T3 maintainer approval. Tip SHA: aa57d50 Ran: - ARCH_DOCS_BASE=origin/main TZ=UTC npm run verify (Node 24.21.0) → pass - node --test .github/scripts/*.test.mjs → pass (88/88) Not run: - smoke / Desktop / live-e2e with a real model — no runtime code changed T3 (touches AGENTS.md and .github/**): needs the maintainer's approval; not merged by the agent.
Part of #384
What changed
bt-row-wrap), the settings gear is absolutely positioned over the row so the hover and selected background fill the same full-width band as a plain row (including at 150% zoom). The row gets a little right padding while the gear is visible so the title does not sit under it.lookAvoidingLocalssteps a friend's Bot seeded shape (then color, once all shapes are taken) when that fallback would match the exact shape+color of one of your own Bots. A named ink and its published lowercase hex count as the same color.indexRosterregisters the own Bots andBotMarkapplies the avoidance to every remote Bot, so sidebar, header, palette, details drawer, room avatars and transcript marks agree. A published image, emoji, or stored shape is left alone (real look sync). Never picks the Main Bot whale.Still open on #384: one short name format in chat header / picker / approval cards (touches #305 alias routing).
Verification
TZ=UTC ARCH_DOCS_BASE=upstream/main npm run verify(Node 22, harness 0.2.0-rc.2) at the tip: passcore/test/looks.test.jscovers clash avoidance, named-ink vs hex, full shape exhaustion, and the roster registryDocs-not-needed: subsystem note updated in docs/subsystems/core-client.md.