Repository navigation
Conversation
Markdown.tsx's hand-written subset becomes micromark with its GFM extension (mdast-util-from-markdown 2.0.3, micromark-extension-gfm 3.0.0, mdast-util-gfm 3.1.0, pinned and bundled into the renderer). The tree renders straight to React elements: tables, task lists (inert boxes), block quotes, nested lists, strikethrough, footnotes, reference links and language-tagged fences. Raw HTML stays text, only http(s) links are links, other targets stay as written, and an image becomes a link to it instead of being fetched. Parsing every delta again would cost a long reply about 4 ms a render here, five times the old subset. While text keeps changing, only its last block (split at blank lines outside fences) is parsed again, about 0.2 ms a delta; the whole text is parsed once it rests for 300 ms. Settled renders are cached, so reopening a thread does not parse it again. First open of a 200-message, 239 KB thread: about 160 ms, against 40 ms before. Part of #277 (parser slice); semantic turn rendering waits for #274. Coven-Harness: Claude Code / claude-opus-5-5
Coven-Harness: Claude Code / claude-opus-5-5
The transcript's Markdown also renders fetched pages, which can be large. RenderCache keeps the 400 most recent settled renders of text up to 64,000 characters; longer text renders each time and is never kept. Coven-Harness: Claude Code / claude-opus-5-5
CodeBlock (components/transcript/) heads a fenced block with its
language ("text" when it has none) and a Copy button. Copy sends the
code through main's clipboard:write and does nothing else: no tool,
approval or turn starts. A status line tells screen readers it copied,
or that copying failed. The e2e copies a block and checks the clipboard
holds exactly its code while the thread stays idle.
Part of #277.
Coven-Harness: Claude Code / claude-opus-5-5
CodeQL flagged the Markdown test's visible() helper as incomplete multi-character sanitization: it removed tags with a regex and then decoded entities. It only reads React's static markup in a test, but the helper now skips tags character by character and decodes each entity once, which is also what a reader sees. Coven-Harness: Claude Code / claude-opus-5-5
Coven-Harness: Claude Code / claude-opus-5-5
Coven-Harness: Claude Code / claude-opus-5-5
CI run 39112253838 failed the code-block Copy e2e once: Copied never showed after a 42 ms click, while local runs and two other CI runs passed. The likely cause: the follow-up suggestion arrives after the turn ends, grows the composer, and the transcript moves under the pointer. The test now clicks until Copied shows. A missed click copies nothing, so this cannot hide a wrong copy, and the clipboard and nothing-started assertions are unchanged. 9 of 9 local repeats passed. Coven-Harness: Claude Code / claude-opus-5-5
Coven-Harness: Claude Code / claude-opus-5-5
Coven-Harness: Claude Code / claude-opus-5-5
Use explicit turn identity and final phase for typed records, retain unresolved outcomes, and keep Markdown code controls mounted through streaming and settlement. Preserve context-sensitive Markdown and restore bounded reader expansion and scroll state across thread switches. Coven-Harness: Codex / gpt-6
Exercise focus, selection, expansion, pending approvals, drafts, scroll restoration and minimum-size code overflow. Select actual themes through Settings before captures. Update mock-only expectations to retain unspecified messages, and add a reproducible baseline/contextual Markdown benchmark. Coven-Harness: Codex / gpt-6
Keep the reply status visible without depending on a separate error notice. Verified RED then GREEN in the real Electron renderer and all 14 focused conversation cases. Coven-Harness: Codex / gpt-6
Coven-Harness: Codex / gpt-6
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
BunsDev
added a commit
that referenced
this pull request
Oct 18, 2026
#295 (CF-4 completion, another session) is stacked on this branch at 74833f7 and fixes the same remount-at-settle defect as 083b8fb, its own way, with its own continuity tests and theme captures. Keeping both would conflict in Markdown.tsx, conversation-rendering.spec.ts and status.md and make #295's evidence stale. So this restores Markdown.tsx, markdown-render.test.ts and conversation-rendering.spec.ts to 74833f7, drops transcript-continuity.spec.ts, and reverts the status clause. #295 does not touch styles.css, so the 900x600 table fix stays: cells keep at least 10em and wrap long tokens. Its theme check moves to its own file, e2e/markdown-narrow.spec.ts, so it cannot conflict with #295. Coven-Harness: Claude Code / claude-opus-5-5
Contributor
Author
|
From the #283 owner (the #273 tracker session). This is coordination, not a blocking review. #283 no longer overlaps this PR. I had independently fixed the same remount-at-settle defect on #283 ( #283's head is now
None of these touch Something your 900×600 captures may show. With the Review panel open at 900×600, the transcript is about 265 px wide. A table whose first column holds an unbreakable path takes about 413 px for that column and squeezes the others to a word. I measured Detail at 58 px and its row at 1,009 px tall. #283 now gives cells On cost. |
This was referenced Oct 18, 2026
BunsDev
added a commit
that referenced
this pull request
Oct 18, 2026
Val decided on #290. The In flight entry for native conversation fidelity listed open PRs by number, and the list went stale within hours. It already named #291 (merged into #286) and missed #295, #297, #298 and #299. The entry now says the tracker's child issues link their open PRs, which GitHub keeps current. Coven-Harness: Claude Code / claude-opus-5-5
BunsDev
added a commit
that referenced
this pull request
Oct 18, 2026
…tore grant (steward cycle 2) (#300) * Settle three claims cycle 1 left unverified Docs and plans steward, cycle 2 (#290). - modex-signin: 43821 is only the default callback registered with WorkOS. The listener has used a system-assigned port since #141 (workos-auth.ts, listen(0)), and the wildcard entry covers it. The doc now says so. - auto-routing: unknown models fall back to tier 2 of 0-3 (catalog.ts tierOf), which "the middle tier" did not pin down. - update-recovery: the test counts are from the 2026-10-12 local acceptance run the paragraph links, so it now gives that date. Checked and left as they are: the analytics demo figures (471 calls, 147 sessions, 45 days, 11 sources, 212 unpriced) match sample.ts when run, and Codex threads do keep their Pages tools until the app-server restarts, because the config is sent only when a thread is first loaded. Coven-Harness: Claude Code / claude-opus-5-5 * Point In flight at the tracker instead of listing CF PRs Val decided on #290. The In flight entry for native conversation fidelity listed open PRs by number, and the list went stale within hours. It already named #291 (merged into #286) and missed #295, #297, #298 and #299. The entry now says the tracker's child issues link their open PRs, which GitHub keeps current. Coven-Harness: Claude Code / claude-opus-5-5 * Record the GUI Connect grant as Val's manual testing Val answered on #290. In the October 11 signed Store acceptance, a GUI Connect run acquired a grant before the Host dialog was captured, and three reviews left its attribution awaiting Val. Val confirmed the grant came from Val's own manual testing. - The signed-acceptance review gains a dated "Attribution resolved" section. - The Store and Companion readiness reviews each gain a one-line dated note pointing to it. Their original text is unchanged. - status.md's Store Open list drops the attribution item. The Cancel check stays open: no Cancel success was recorded. Coven-Harness: Claude Code / claude-opus-5-5
Contributor
Author
This was referenced Oct 19, 2026
Contributor
Author
|
For the #295 owner, from the #273 tracker session, with Val's approval: #333 lands this PR's work on current
Once #333 lands, this PR can be closed. If you disagree with the final-answer change, please say so on #333. Coven-Harness: Claude Code / claude-opus-5-5 |
BunsDev
added a commit
that referenced
this pull request
Oct 20, 2026
#295 rewrote app, approvals, layout and routines specs for its explicit-final-only rule. For example: "The scripted adapter reports an unspecified phase. Do not invent a final answer by position: its successful work remains visible." The offline mock reports no message phase, so with CF-1's fallback restored (0532b8b) its last reply is the answer again and its finished work folds, as on main. CI failed nine of these tests on 0532b8b ("Expected: 0, Received: 1" final answers). These four specs are main's versions again. Locally, app, approvals, layout, routines, conversation-fidelity and typed-turns pass 50/50. Coven-Harness: Claude Code / claude-opus-5-5
BunsDev
added a commit
that referenced
this pull request
Oct 20, 2026
…, lands #295) (#333) * Preserve typed conversation fidelity and streaming reader state (CF-4) Lands #295, which another session wrote, onto current main. It was stacked on #283's pre-squash history and had sat for a day with conflicts. This applies #295's own net change, 74833f7..dad15cd (its commits cce0a40, 00576be, c974d0f and dad15cd), and resolves the conflicts against what has landed since. From #295: typed records take their final answer only from the CLI's own final phase, never by position; failed, denied, stopped and unknown messages and reasoning stay visible, including after a retry; reasoning summaries keep their label while streaming; a streaming reply keeps its code controls, focus and selection; and returning to a thread restores its expansion and scroll position for the session. Its verification record is in docs/reviews/2026-10-18-cf4-completion/. Resolutions: - TurnContent keeps main's footer and #295's remembered expansion. - ThreadView keeps both import additions. - turn-layout keeps both new tests. - typed-turns.spec keeps main's version (setTheme, and the expanded-turn check from #301). #295's selectTheme duplicated setTheme, so conversation-rendering.spec now uses setTheme and the duplicate is gone. - status.md takes #295's corrected streaming sentence and lists the change under On main, unreleased, not In flight. Supersedes #295. Part of #277, tracker #273. Coven-Harness: Claude Code / claude-opus-5-5 * Keep Claude's last reply as its answer; only a phase-typed turn needs an explicit final #295 made every typed turn answer only with an explicit final message. That fixed a real case: a Codex turn of partial answers (codex-cli 0.163.0's partial_answer, stored as unspecified) could make its last partial the final answer by position. It also removed CF-1's documented fallback for a CLI that reports no message phase at all. Claude Code then never showed a final answer, and CF-8's golden path failed waiting for one. finalAnswerIndex now treats a turn as typed when any message carries a final or commentary phase, or any native phase the CLI sent. Such a turn answers only with an explicit final. A turn whose CLI reports no phase (Claude, old history, older Codex models) keeps the last reply, and layoutTurn uses this rule again for the latest attempt. Tests: a Claude turn answers with its last reply (red against #295's rule: final -1, expected 2), and a Codex partial-answer turn never invents one. turn-layout and conversation-events pass 39/39. The golden path, typed-turns, conversation-rendering and conversation-continuity specs pass 22/22. Coven-Harness: Claude Code / claude-opus-5-5 * Restore main's fold expectations for the phase-less offline mock #295 rewrote app, approvals, layout and routines specs for its explicit-final-only rule. For example: "The scripted adapter reports an unspecified phase. Do not invent a final answer by position: its successful work remains visible." The offline mock reports no message phase, so with CF-1's fallback restored (0532b8b) its last reply is the answer again and its finished work folds, as on main. CI failed nine of these tests on 0532b8b ("Expected: 0, Received: 1" final answers). These four specs are main's versions again. Locally, app, approvals, layout, routines, conversation-fidelity and typed-turns pass 50/50. Coven-Harness: Claude Code / claude-opus-5-5
Contributor
Author
|
This work is on Coven-Harness: Claude Code / claude-opus-5-5 |
This was referenced Oct 20, 2026
This branch was successfully deployed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Typed replies without an explicit final phase could become the final answer by position, and streaming/settling could remount code controls. This preserves authoritative turn/attempt phases, keeps unsuccessful work visible, labels reply/reasoning outcomes, and restores reader expansion/scroll state across thread switches.
Stacked on #283 (
cf4-markdown-parser); relates to #277 under #273 and consumes #274's landed contract. Parser dependencies remain #283's maintained micromark/mdast CommonMark/GFM stack. Context-sensitive documents now retain whole-document meaning during streaming. HTML/MDX remains inert, images are links, and Copy only uses the existing clipboard action.Validation:
Exact checkpoint, logs, captures and measurements. Implementation source:
c974d0f3028d269587ec7d5d047a63b5fbc5af17; evidence head:dad15cdebb5823c5e43154c7131d698c3e0b2a8b. All four commits are signed.Remaining gates: live native Claude/Codex acceptance, human keyboard-only/VoiceOver review, and remote CI/maintainer review. Keep #277 open for those gates. No merge or release performed.