apps/console/docs/UI_IMPROVEMENT_PROPOSAL.md:167 documents a direction prop on ResizablePanelGroup — the exported component takes orientation #7282

Activity

1. os-project-manager commented on Sep 2, 2026

   

   os-project-manager

   on Sep 2, 2026 – with Claude

   Collaborator

   More actions

   Graded under the objectui#6600 ruling — documentation · domain:ui · pm:queue; scope widened to the fragment markers

   Director seat (objectstack #12708), summon #8, session session_01ShyhexkB2d1AeRZ85tgAAe, 2026-09-02.

   Provenance: maintainer ruling D on objectui#6600 (decision batch #8, 2026-09-02, verbatim 「同意」): fix the content first, then move the documentation gates' roots. This card is step 1 of that ruling, and #6600 is pm:blocked on it. Filed unlabelled by the measuring seat; graded here so the ordering the maintainer ruled has a card to run on.

   Scope (all in apps/console/docs/**, ui-lane content):

   1. Decide first whether UI_IMPROVEMENT_PROPOSAL.md is still a live document. If it is not, retire the file, which closes the prop defect with it. If it is, fix the direction prop at :165 to orientation (packages/components/src/ui/resizable.tsx:44 is the declaration; resizable-orientation.test.tsx and navigation-overlay.tsx:519 are the in-tree references).
   2. Mark every excerpt-fragment ts / tsx block in the surviving guides with the check-doc-snippet-types.mjs gate's per-block doc-snippet: fragment marker, each with a written reason. From the finding(gate): apps/console/docs/** is outside every documentation gate's scan surface — the console's operator guides are checked by nothing but check:control-bytes #6600 measurement: 14 compiled blocks across the tree, 11 red, of which this card's TS2322 is the only defect in what the prose claims; the rest are fragments (undeclared identifiers, one unclosed JSX element). No UNGATED_DOCS entries: the ruling rejects whole-file exemptions.
   3. Docs-only change ⇒ empty-frontmatter changeset per objectui#7089 §3 (the implementer confirms apps/console is private: true before choosing that form).

   Acceptance: with the two gate roots pointed at apps/*/docs/** in a scratch run (the method the #6600 measurement used, the gates' own exported functions), check:doc-snippets reports 0 failed for this tree. That is the condition #6600's step 2 needs before it moves the roots for real.

   Clause-② no. S. When this card closes, #6600 unblocks.

   ---

   Generated by Claude Code
2. os-project-manager

   added

   documentationImprovements or additions to documentationImprovements or additions to documentation

   pm:queue

   domain:uiobjectui ui stream: fix lands on the published library or apps — objectui execution seatobjectui ui stream: fix lands on the published library or apps — objectui execution seat

   pm:dispatched

   and removed

   pm:queue

   on Sep 2, 2026
3. os-project-manager

   self-assigned this

   on Sep 3, 2026
4. os-project-manager commented on Sep 3, 2026

   

   os-project-manager

   on Sep 3, 2026 – with Claude

   Collaborator

   More actions

   Claim: domain:ui execution seat, session session_01EMrWaQw3XS5DxTHxp4yRyC, branch claude/issue-7282-console-docs-fragment-markers.

   Dispatched to one os-dev agent this round, against the director seat's grading in comment 5510095526 (ruling D on #6600, 2026-09-02, verbatim 「同意」) rather than against the body alone — the grading is what carries the acceptance criterion and steps 2 and 3, which the body does not.

   File surface declared: apps/console/docs/** and a changeset. No packages/** and no gate script is in scope — moving the gate roots is #6600's step 2, which this card unblocks rather than performs.

   One instruction added on top of the grading, stated so it is visible rather than folded in: step 1 offers retirement of UI_IMPROVEMENT_PROPOSAL.md as an alternative to fixing the prop. Deleting a file is not a coin-flip, so the dispatch requires that the fix branch be taken unless the evidence for retirement is decisive, with the liveness evidence recorded in the PR body either way. If the implementer reads the evidence as decisive for retirement, it reports rather than deletes and the question comes back here.

   ---

   Generated by Claude Code
5. os-project-manager commented on Sep 3, 2026

   

   os-project-manager

   on Sep 3, 2026 – with Claude

   Collaborator

   More actions

   os-dev-report

   {
     "issue": 7282,
     "status": "done",
     "branch": "claude/issue-7282-console-docs-fragment-markers",
     "pr": "https://github.com/objectstack-ai/objectui/pull/7425",
     "premise_still_valid": true,
     "summary": "Took the FIX branch, not retirement: UI_IMPROVEMENT_PROPOSAL.md is live. Evidence for — apps/console/README.md:134 links it from the app's Documentation table; its subject is live code (packages/app-shell/src/views/MetadataInspector.tsx, consumed at apps/console/src/preview-gallery.tsx:51,110 via getMetadataInspector); the API it teaches is a current export (packages/components/src/ui/resizable.tsx:44, pinned by resizable-orientation.test.tsx:52, called at navigation-overlay.tsx:519); objectui#4938 deliberately brought apps/*/docs/** under check-doc-links and that gate's test names this exact path. Evidence against, recorded and not decisive — created 2026-02-11 in 366775d42 (copilot-swe-agent, PR #444, merged 65901a5ed) and never modified since, and it names one stale path (apps/console/src/components/MetadataInspector.tsx, which moved to packages/app-shell/src/views/). WARNING for anyone re-checking: the container clone is SHALLOW, and there git log -- path misattributes the file to 835a2cebe (a repo-wide lint fix, PR #5978) which in full history does not touch it; I unshallowed before reading the date. Landed: the direction prop corrected to orientation at :168; ten per-block doc-snippet fragment declarations with written reasons naming the identifiers each excerpt leaves to its host (8 in UI_IMPROVEMENT_PROPOSAL.md, 2 in error-tracking.md); no UNGATED_DOCS entry; an empty-frontmatter changeset. TWO PREMISE CORRECTIONS FOR PM. (1) The grading's step 3 says to confirm apps/console is private: true — it is NOT. @object-ui/console has no private key and is a published member of the 40-package fixed group at 17.6.0. The empty-frontmatter form is still correct, but on a different verified ground: apps/console/docs/** is not in the package's files (dist, plugin.ts, plugin.js, plugin.d.ts, README.md), so no published artifact moves, and check-changeset-presence.mjs independently reports no changeset owed. (2) SELF-CORRECTION, caught in PM review: my first report claimed line-number drift in the card's references, saying the in-tree call site was navigation-overlay.tsx:520 rather than :519. That claim was WRONG and the card was right — git grep -n at bf244f400 puts the ResizablePanelGroup call site at :519, and :520 is the child ResizablePanel. The bad figure came from counting off a sed line range by hand instead of asking grep -n. It had reached the durable changeset as well as the PR body; commit 56fab342e corrects the changeset to :519, the drift claim is removed from the PR body entirely rather than inverted, and all three of the card's references (resizable.tsx:44, resizable-orientation.test.tsx:52, navigation-overlay.tsx:519) are verified exact. The commit message of 3b067b9b0 still carries the wrong figure and cannot be amended without a force-push; 56fab342e's message states the correction so the record is self-correcting. No gate script touched and no gate root moved.",
     "census": {
       "method": "the gate's own exported scanFences + compileSnippets, measured on bf244f400, roots pointed at apps/*/docs/** by a scratch harness that replays analyze()'s glue; nothing in the repo modified",
       "ts_tsx_blocks_total": 14,
       "red_before": 11,
       "red_before_breakdown": "2 syntax (one unclosed JSX div at :64, one bare object literal at :185) + 9 semantic",
       "defects_in_what_the_prose_claims": 1,
       "excerpt_fragments": 10,
       "already_green": 3,
       "per_document": "UI_IMPROVEMENT_PROPOSAL.md 12 blocks (9 red); error-tracking.md 2 blocks (2 red); deployment.md 0 blocks",
       "matches_card": "yes — the card's 14/11 reproduce exactly on current main",
       "after": "14 blocks — 4 to compile, 10 declared fragments, 0 failed",
       "note_on_the_three_left_undeclared": "they are the design-token lists under Visual Design Specs; they parse as labelled statements and compile clean, so a fragment marker would assert something false and would drop three blocks out of real semantic coverage. Filed as objectui#7426 instead.",
       "note_on_card_references": "all three verified exact at bf244f400 with git grep -n: resizable.tsx:44, resizable-orientation.test.tsx:52, navigation-overlay.tsx:519. An earlier claim of drift at :520 was my own error and is retracted."
     },
     "tests": "ACCEPTANCE (the ruling's own criterion), scratch run with the two gate roots pointed at apps/*/docs/**, all four controls held — resolution landed on packages/types/dist/index.d.ts, sentinel 1 diagnostic TS2305, positive 0, undeclared @floating-ui/react-dom 1 diagnostic TS2307: 'Scanned 3 document(s): 3 covered (2 of them hold a ts/tsx block), 0 ungated. Covered blocks: 14 - 4 to compile, 10 declared fragment(s). Syntax phase: every block parsed. Semantic phase: 4 of 4 block(s) judged, 0 failed.' RAW_EXIT=0. check:doc-snippets reports 0 failed for this tree, which is #6600 step 2's precondition. REVERSE VERIFICATION, direction predicted as red before running, fix committed FIRST so each leg restores from HEAD; every mutation proven on disk by counting injected and removed text, every restore proven by hash equality with the HEAD blob plus an empty git diff HEAD (the gate reads the markdown from disk, so no package rebuild is part of the mutation; the dist the types come from was built and unchanged throughout). Leg A, revert the fix to direction=: RED, 1 failed, TS2322 on that block. Leg B, plant orientationX=: RED, 1 failed, and TypeScript's own diagnostic ends 'Did you mean orientation?' — proving the block's props are genuinely excess-checked rather than widened to any. Restored control: 4 of 4 judged, 0 failed, tree byte-identical to HEAD. GATE UNION re-run AFTER the final commit, at 3b067b9b0, clean tree, each exit code captured before any pipe and each verdict quoted from the gate's own line: docs:check-links 0 ('Links are valid across 17 scan roots'); check:doc-snippets 0 ('Every covered documentation snippet compiles against the built types', 417 of 417 judged 0 failed) — NOTE it first returned exit 2, PRECONDITION NOT MET, the gate's own could-not-run code and NOT a documents verdict, so its 25-package --build-filter closure was built and it was re-run, and the 0 is that run; check:doc-types 0; check:doc-fences 0; check:control-bytes 0 (6111 files); check-changeset-presence.mjs 0 (no changeset owed); check-changeset-no-major.mjs 0. Heavy builds went through scripts/pm/os-verify-lock.sh (VERDICT command-exit 0 both times). NARROWING DECLARED: pnpm lint is a repo-wide scan CI owns and was not run whole. The narrowing is a measurement, not a skip — the population came from eslint's OWN config resolution, which reports all three changed files as 'File ignored because no matching configuration was supplied' with 0 errors under --format json; the flat config carries no .md matcher; and the diff adds no config, no files pattern and no source file, so no untouched file's verdict can move. Control-byte self-scan on the three changed files: no matches. FOLLOW-UP after PM review: commit 56fab342e changes two tokens of changeset prose and nothing else (git diff --stat 3b067b9b0 HEAD = 1 file, 1 insertion, 1 deletion; working tree clean). At the reviewer's direction the full union was NOT re-run for a text-only edit; pnpm check:control-bytes (RAW_EXIT=0, 6111 files) and check-changeset-presence.mjs (RAW_EXIT=0, no changeset owed) were re-run on the new head. The census, the fix, the markers and the acceptance run are untouched by that commit, so the figures above still describe the tree at head 56fab342e.",
     "mcp_calls": "5 — search_issues (dedupe, one targeted call), create_pull_request, issue_write (#7426), update_pull_request (the review correction), add_issue_comment (this report). CHANNEL SWITCH DECLARED: repo-scoped REST reads work in this container and carried the issue, its comments, and every PR body read-back at zero quota, but the REST SEARCH endpoint is refused with HTTP 403 and the message 'sessions are bound to their configured repositories, use repository-scoped endpoints', so the dedupe went through one targeted MCP search_issues. That search returned 14 results, so it was not a vacuous empty.",
     "open_questions": [],
     "out_of_scope_findings": [
       "filed as #7426: three tsx fences in apps/console/docs/UI_IMPROVEMENT_PROPOSAL.md (Visual Design Specs, :199 :207 :215) are design-token lists that compile only as labelled statements, so once #6600 step 2 moves the roots three of this tree's four live compiled blocks are a check that cannot fail. Same class as the already-closed #5997 one tree over, which was fixed by re-fencing; the fence target here needs deciding (not valid JSON, not a bullet list). Left unfixed on purpose: re-fencing is a document-shape change of a different class from this card's defect.",
       "not filed, recorded in the PR body only: UI_IMPROVEMENT_PROPOSAL.md still names apps/console/src/components/MetadataInspector.tsx, a path that no longer exists — the component moved to packages/app-shell/src/views/. Rewriting a design proposal's prose is outside this card."
     ],
     "head": "56fab342e"
   }

   ---

   Generated by Claude Code
6. os-project-manager commented on Sep 3, 2026

   

   os-project-manager

   on Sep 3, 2026 – with Claude

   Collaborator

   More actions

   ✅ ACCEPT — PR #7425 at 56fab342e. Enqueued 01:35Z.

   domain:ui execution seat, session session_01EMrWaQw3XS5DxTHxp4yRyC. Every check green on the corrected head — 26 success, 3 skipped, 0 failed, 0 running, not the required subset — flipped out of draft and enqueued; enqueue confirmed by the queue ref pr-7425-42c129b64… (suffix = main, so it is first in line). Clause ② does not apply, per the grading's "Clause-② no. S."

   The ruling's own acceptance criterion is met

   A scratch harness replaying the gates' exported functions over apps/*/docs/** reports 4 of 4 blocks judged, 0 failed, with four controls held (resolution landed on packages/types/dist/index.d.ts; sentinel TS2305; positive 0; undeclared specifier TS2307). That is #6600 step 2's precondition, so #6600 is unblocked by this landing.

   Verified independently, not taken from the report

   • The correction diff is exactly what was asked: git diff --stat 3b067b9b0 56fab342e = 1 file, 1 insertion, 1 deletion, and it is the single token :520 → :519 in the changeset. navigation-overlay.tsx:520 now has 0 occurrences anywhere in the tree; :519 is present.
   • The census reproduces. Counting ts/tsx fences in UI_IMPROVEMENT_PROPOSAL.md on origin/main myself gives 12, matching the per-document figure in the PR body.
   • Pre-merge baseline taken on 42c129b64 for post-landing verification: direction="horizontal" 1, orientation="horizontal" 0, doc-snippet: fragment 0; controls ResizablePanelGroup 3 and 12 fences must survive on both sides.

   ⭐ Two things the implementer got right that are worth recording

   1. It falsified this card's own grading. Step 3 of the grading instructed the implementer to "confirm apps/console is private: true" before choosing the empty-frontmatter changeset. It is not: @object-ui/console has no private key and is a published member of the 40-package fixed group at 17.6.0. Rather than satisfying the instruction as written or silently doing the right thing, it re-grounded the declaration on a fact it verified — apps/console/docs/** is not in the package's files — and disclosed the divergence. ⚠️ Flagging for the director seat: that false premise is in a grading comment and could be reused on the next apps/* card.

   2. It retracted rather than inverted. The first submission claimed to have found line-number drift in this card's references (:520 rather than :519). That claim was wrong — the card was right, :520 is the child ResizablePanel — and it had reached the durable changeset. Asked to fix it, the implementer removed the claim entirely instead of writing a corrected sentence saying the card was right, which would have been noise; and it declined to force-push to scrub the stale figure from 3b067b9b0's commit message, letting 56fab342e's message carry the correction so the history is self-correcting rather than rewritten. ⛔ Never rewriting pushed history was the right call and was not something the dispatch specified.

   Carried forward, open

   • finding(docs): three tsx fences in apps/console/docs are design-token lists that compile only as labelled statements — the snippet gate will be green about nothing once #6600 moves its roots #7426 — the three "Visual Design Specs" token-list blocks compile only as labelled statements, so once finding(gate): apps/console/docs/** is outside every documentation gate's scan surface — the console's operator guides are checked by nothing but check:control-bytes #6600 step 2 moves the roots, three of this tree's four live compiled blocks become a check that cannot fail. ⚠️ Deliberately left undeclared here rather than marked fragment: they are not fragments, and a marker would assert something false while dropping them out of real coverage. Same class as the already-closed Six tsx-fenced blocks in guide/component-registry.md are markdown bullet lists, not TypeScript — re-fence them and drop the six FRAGMENT_MARKERs #5991 had to declare #5997; the fence target needs deciding.
   • Recorded in the PR body only: UI_IMPROVEMENT_PROPOSAL.md still names apps/console/src/components/MetadataInspector.tsx, which moved to packages/app-shell/src/views/. Rewriting a design proposal's prose is outside this card.

   ---

   Generated by Claude Code
7. os-project-manager

   closed this as completedin #7425on Sep 3, 2026
8. os-project-manager

   removed

   pm:dispatched

   on Sep 3, 2026