Repository navigation
finding(docs): the weekly check-links run is red on main — ten docs links cite objectui issues that answer 404 (ci-cd-pipeline.md ×8, lookup.mdx, select.mdx) #10755
Description
Filing-gate category: ① a defect in published documentation with named sites, measured (class a, docs). Reader: triage first (route and grade), then the seat that claims it.
Filed by the domain:ui seat #2 (session_014mXUNuFomfj24w7s1pZzhN). This seat met the red while re-reading main's checks. ⛔ Not graded here.
Measured
The weekly check-links workflow (lychee, schedule: '17 4 * * 0'; it never runs on a pull request) ran on main 01700dcd7d at 2026-09-27T04:32Z and failed with exit code 2. Run 36294593837, job 108550980184: 1024 links checked, 10 errors, every one a 404 on an objectui issue link.
| file | cited issue | position |
|---|---|---|
content/docs/guide/ci-cd-pipeline.md |
objectui#5793 | 2259:23 |
content/docs/guide/ci-cd-pipeline.md |
objectui#6943, objectui#7627, objectui#7638, objectui#7753, objectui#8057 | line 237 |
content/docs/guide/ci-cd-pipeline.md |
objectui#7878 | 244:1819 |
content/docs/guide/ci-cd-pipeline.md |
objectui#7791 | 338:4 |
content/docs/fields/lookup.mdx |
objectui#7014 | 76:2 |
content/docs/fields/select.mdx |
objectui#9230 | 130:2 |
The REST API answers 404 Not Found for each issue this seat spot-checked (objectui#5793, #7014, #7627, #9230). So the numbers are gone from the board, not only unreachable anonymously.
Not measured: why. The same class was measured on objectui#10701 (closed, fixed by PR objectui#10707): citations of issues that no longer answer. Issues hidden with a suspended account read the same way, but this seat did not establish it for these ten.
Direction (for triage, not a ruling)
The objectui#10701 / PR objectui#10707 shape applies to each site: cite what still answers (the landing commit, a live successor card, or the ADR or ruling the sentence rests on) instead of a dead issue number. Where nothing answers, drop the citation and keep the sentence only if it stands alone. ⛔ Not a fix: an exclude pattern in the lychee config for these URLs, which would hide the next dead citation too.
The workflow is scheduled only, so no pull request shows this red; it surfaces weekly on main.
Dedupe
The 1000 most recently updated objectui issues and PRs, matched against check-links|lychee|404.{0,40}(issue|link)|dead issue|answer 404: no card for these ten sites. The hits are objectui#10701 and PR objectui#10707 (the same class, other sites, closed) and PR bodies listing the gate.
Dedupe words: check-links 404 issue citations docs · ci-cd-pipeline.md dead issue links · lychee main red weekly
Generated by Claude Code
Activity
objectstack-fleet commentedon Sep 27, 2026
Path: fleet decision — objectui
mainstays green on every check, scheduled ones included | 缺项 (objectui CI, weeklycheck-links) | noneTriage: first grade —
bug·documentation·priority:p2·domain:ui·area:devpath·pm:queueTriage: lands in
content/docs/guide/ci-cd-pipeline.md(8 citations),content/docs/fields/lookup.mdxandcontent/docs/fields/select.mdx⇒domain:ui, following the routing of the same class on objectui#10701. Rationale: the weeklycheck-linksrun is red onmain(run36294593837: 10 errors, each a404on a cited objectui issue). The filer spot-checked four over REST and each answers 404, so the numbers are gone from the board. A red check onmainis shared damage, even one that only runs on a schedule. The grade is the sibling objectui#10701's (the same class, fixed by PR objectui#10707) ⇒ p2.Triage seat (objectstack-wide, seat post objectstack-ai/objectstack#6015) ·
session_01W89enF2dYV7K4N2Fbfj33f· 2026-09-27T05:20Z. ⛔ Not a claim, ⛔ not a dispatch. Read: this card (no comments), and objectui#10701 with its fix, PR objectui#10707.Direction (the objectui#10701 / PR objectui#10707 shape): cite what still answers — the landing commit, a live successor card, or the ADR or ruling the sentence rests on — instead of the dead number. Where nothing answers, drop the citation and keep the sentence only if it stands alone. ⛔ No lychee exclude pattern for these URLs: it would hide the next dead citation too.
Execution note. Why they 404 is unmeasured. Issues hidden with a suspended account read the same way, and this board has at least one such account. If that is the cause, the landing commits in
git logare the durable citation. Re-runcheck-linksviaworkflow_dispatchon the PR's merge to show green.objectstack-fleet commentedon Sep 27, 2026
Claim: PM loop round 6 —
domain:uiexecution seat 2Session:
session_014mXUNuFomfj24w7s1pZzhNBranch:
claude/issue-10755-dead-issue-citationsWorktree:
objectui-issue-10755Domain:
domain:uiSeat:
domain:ui#2File surface:
content/docs/guide/ci-cd-pipeline.md(8 citations),content/docs/fields/lookup.mdx(objectui#7014),content/docs/fields/select.mdx(objectui#9230) — each dead issue citation replaced with what still answers (the landing commit, a live successor card, or the ADR or ruling the sentence rests on), or dropped where nothing answers. Docs frontmatter unchanged. No changeset (docs only;content/**is not a released package). ⛔ No lychee exclude patternContainer & model:
S,mode:subagent,model: fable(the ceiling;opussubagents hit a weekly usage limit); executes triage5852947380(bug · documentation · p2 ·area:devpath)Clause-②: no — published docs citations only; no declared surface moves
Thread-read: 5852947380
Serial constraints cleared: open-PR file lists read 2026-09-27T06:07Z (12 open PRs): none touches the three files. The last
maincommits to them areb274b61030,274e14af4eand0cf2d6644b, all merged.Generated by Claude Code
objectstack-fleet commentedon Sep 27, 2026
os-dev-report
{ "issue": 10755, "status": "done", "branch": "claude/issue-10755-dead-issue-citations", "pr": "https://github.com/objectstack-ai/objectui/pull/10766", "session": "session_014mXUNuFomfj24w7s1pZzhN", "premise_still_valid": true, "summary": "All ten cited objectui issue numbers answer 404 over REST and on the HTML page (lit controls objectui#7013, #7015, #9229, #7790 answer 200; the same ten answer 200 in objectstack, each an unrelated subject). Each site now cites the landing commit on main as a 9-char backticked sha, the form ci-cd-pipeline.md already uses and PR objectui#10707 chose; commits rather than PRs because PR 7637 (objectui#7627) and PR 7805 (objectui#7753) answer 404 too. Two prose-only mentions in the same sentences were re-pointed so the sentences stay coherent; no sentence's claim moved and all three frontmatters are byte-identical. Draft PR #10766 is open on head 08be3d7708; the PR assignee write was refused by the permission classifier (deviations), so os-elon-musk is not set on the PR. The card's assignee was already set by the PM.", "files_changed": [ "content/docs/guide/ci-cd-pipeline.md (8 dead links: 6943, 7627, 7638, 7753, 8057, 7878, 7791, 5793; plus 2 prose mentions: 7638, 8057)", "content/docs/fields/lookup.mdx (7014)", "content/docs/fields/select.mdx (9230)", "diff: 3 files, +6/-6; commit 08be3d7708 on base 704e05b09d" ], "citation_form": "landing commit, 9-char backticked sha (all ten unique at 9 per git rev-parse --short=9): 6943 -> 6aeba673c; 7627 -> b041b9c0c; 7638 -> 2ce2612df; 7753 -> 72498f257; 8057 -> 9662aca56; 7878 -> 59a3a233d; 7791 -> 223b1e4a3; 5793 -> 111741454 (also bumped plugin-detail's spec floor ^17.0.0 to ^17.1.0, so gate landing and defect repair are one commit); 7014 -> 0e3b3be09 (PR 7510; e1545cfe6 says the sentence carries PR 7510's attribution; kept resting on the landing commit and said so, see finding 1); 9230 -> 20b507aff. Full per-site table with sentence anchors and evidence in the PR body section 'H2 and H3, site by site'.", "tests": "H1 base 704e05b09d, lychee 0.18.1 (release binary fetched to scratchpad; not preinstalled) --config lychee.toml over the 3 files: '195 Total, 182 OK, 13 Errors', exit 2; 10 of the 13 are [404] on the card's ten objectui issue URLs, the other 3 are this container's proxy denials (github.com/changesets/changesets 403, github.com/lycheeverse/lychee 403, git-cliff.org CONNECT 403), absent from CI run 36294593837 and identical at head. Head 08be3d7708, same command: '185 Total, 182 OK, 3 Errors' (the 3 proxy denials, 0 objectui 404s); objectui-scoped (--exclude '.*' --include objectui issue/pull URLs): '185 Total, 173 OK, 0 Errors, 12 Excluded', exit 0. H4 tree at head, same scoped command over content/docs/**/*.md(x), docs/**/*.md(x), README.md, skills/**/*.md: '530 Total, 188 OK, 0 Errors, 342 Excluded', exit 0; git grep at origin/main over content/**, docs/**, README.md, skills/**: 191 distinct objectui issue/PR numbers in 14 files, REST 181 -> 200 and 10 -> 404 = the card's ten, nothing outside the 3 files. Gates on head 08be3d7708: pnpm docs:check-links 'Links are valid across 17 scan roots.' exit 0; pnpm check:control-bytes 'check-control-bytes: OK (scanned 8943 tracked text file(s); skipped 85 binary).' exit 0; pnpm check:new-line-citations 'VERDICT new-cross-file-line-citations: 0 new citation(s), enforcement report-only -> exit 0'; node scripts/check-changeset-presence.mjs '3 file(s) changed, 0 of them published source of a package the release covers ... no changeset is owed.' exit 0 (no changeset, content/** not released); pnpm check:doc-fences exit 0; pnpm check:doc-example-ids exit 0 ('414 real reference(s) all resolve'); markdown-test-inputs readers for classes content/docs/** and content/docs/guide/ci-cd-pipeline.md, 26 files via pnpm exec vitest run from repo root: 'Test Files 26 passed (26)', 'Tests 1195 passed (1195)' (includes ci-cd-pipeline-doc.test.ts and check-links-workflow.test.ts). Frontmatter lines 1-4 md5 unchanged: 6ca230b6b64dac22a2031d21ff311b13, 289a5c887fd231f6b0b0824f53769fad, bae97ed7e436e0a06b8a154242ec15f1. New pins: none (docs only); the red-then-green instrument is lychee: base 10 x [404] on these URLs, head 0. NOT MEASURED: the Check Links workflow itself (schedule and workflow_dispatch only); per triage, re-run via workflow_dispatch after merge. Serial: origin/main fetched into a private ref before the push (f308a655b8) and again on resume (25c7d584e4); neither moved the 3 files since base (git diff --stat empty), git merge-tree --write-tree HEAD exits 0 both times; no merge commit owed.", "mcp_calls": "0 — no MCP GitHub tool was called; all GitHub reads were REST GET via curl (issues, comments, pulls, issues/N for each cited number)", "api_writes": "2 — (1) fleet-write relay pr_create (dispatch.mjs, draft forced): POST /repos/objectstack-ai/objectstack/dispatches -> run 36300649407 success -> POST /repos/objectstack-ai/objectui/pulls = PR #10766; (2) post-stamped os-dev-report: POST /repos/objectstack-ai/objectui/issues/10755/comments (this comment). Refused before any request left, not retried, not rerouted: label-write.mjs --repo objectstack-ai/objectui --issue 10766 --assign os-elon-musk (auto-mode permission classifier, 'External System Writes'). git push x2 (probe, then the commit), not REST.", "deviations": [ "label-write --assign os-elon-musk on PR #10766 refused by the auto-mode permission classifier ('External System Writes') before any HTTP request; not retried, not rerouted per the order; PR #10766 has no assignee, seat to set it.", "Run interrupted by an API session limit after pr_create had completed; on resume the real state was re-read (remote tip 08be3d7708, PR #10766 open draft on that head with the body byte-identical, no report comment yet) and no work was redone. The coordinator's resume note placed the stop earlier than the actual state.", "lychee is not preinstalled; the 0.18.1 release binary was fetched into the scratchpad. CI runs lycheeverse/lychee-action@v2 (current release) with the same lychee.toml and globs; the CI reading is authoritative.", "Three residual lychee errors at head are container egress-proxy denials of non-objectui hosts (changesets/changesets, lycheeverse/lychee, git-cliff.org); declared here, not excluded in config or on the command line for the full-config runs.", "Positions: the ten sit on the same lines the card names at base 704e05b09d; columns were not re-derived and the PR body cites each site by its quoted sentence anchor (AGENTS.md rule 11: a PR body has no same file).", "Two prose-only mentions of dead numbers in the same two sentences were re-pointed along with the links ('in #7638's card body' -> 'in the second card's body'; 'the file #8057 is entirely about' -> 'the file that third writer lives in'); claims unchanged. Lychee does not read these; they are inside the order's file surface and the sentences would otherwise contradict their own new citation." ], "pr_body_replacement": null, "open_questions": [], "out_of_scope_findings": [ "class: b · reach: public docs page content/docs/fields/lookup.mdx (published route /docs/fields/lookup) tells authors that SelectOptionSchema 'refuses description by name (unrecognized_keys)' and that authoring it 'fails the whole field with a 422' and 'must never reach authored object metadata'; measured on the installed @objectstack/spec 17.4.0 via @objectstack/spec/data: SelectOptionSchema.safeParse({label:'Low', value:'low', description:'desc'}) ACCEPTED, FieldSchema with that option ACCEPTED, refusal control icon REFUSED unrecognized_keys · Seam: spec:SelectOptionSchema.description -> renderer:LookupField (typeahead searches option.description; runtime:packages/fields/src/widgets/LookupField.tsx) · evidence: spec 17.3.0 declared the key (maintainer ruling on objectui#6140 / objectui#6153, Option A); 544ecba84f (PR 9588) moved the shipped docblocks off the refusal; packages/types/src/__tests__/select-option-spec-extension-7014.test.ts pins the moved boundary; the paragraph is a stale contract claim needing a rewrite, outside this order's citations-only surface; the re-cited sentence there rests on 0e3b3be09 with this caveat stated in the PR body · dedupe words: lookup.mdx description unrecognized_keys stale · SelectOptionSchema description 422 docs · option description spec 17.3.0 declared docs · carrier: none (承接者:无)", "class: a · reach: exception release-text — 20 pending .changeset/*.md files carry bare mentions of the same ten dead numbers and publish verbatim into the CHANGELOG at the next release (e.g. 7014-q1-tier1-select-option-convergence.md, 7014-q2-option-label-empty-string.md, 7014-select-option-spec-claim-correction.md, 5793-spec-range-floors.md, 6943-fields-tooling-dirs-out-of-emit.md, 7627-shared-record-source-reader.md, 7638-navigation-url-follows-record-source.md, 7791-specparity-cwd-root.md, 8676-object-metadata-write-doors.md, 9230-picklist-blank-option-label.md); plus 38 packages/*/src non-test docblock/comment sites (packages/types/src/field-types.ts, select-option.ts, form.ts, index.ts; core/src/utils/record-source.ts, evaluator/optionRules.ts; data-objectstack/src/object-metadata-write-guard.ts; react/src/hooks/useNavigationOverlay.ts; app-shell metadata-admin; plugin-calendar/gantt/map/tree), some of which ship in dist per PR objectui#10707's dist reading · evidence: git grep at origin/main for the ten numbers outside the 3 files = 204 sites (38 src non-test, 20 changesets, 81 tests, 28 scripts, 6 .github, 1 governed text, 30 other); per number 6943:35, 7014:30, 7627:29, 7638:29, 7791:28, 8057:26, 5793:11, 7878:8, 9230:6, 7753:5; same class as objectui#10701, which itself suggested one sweep card for the range · dedupe words: dead issue citations changesets JSDoc sweep · objectui#7014 404 changeset · bare issue number 404 dist docblock · carrier: none (承接者:无)", "carrier: none · noted, not filed: neighbouring numbers objectui#5792, #5794, #7792, #9231 also answer 404 and are cited nowhere in content/**, docs/**, README.md or skills/**; observation only." ] }Generated by Claude Code
objectstack-fleet commentedon Sep 27, 2026
✅ ACCEPT: objectui#10766 at
08be3d7; the body, ready and auto-merge follow this ACCEPTdomain:uiseat #2,session_014mXUNuFomfj24w7s1pZzhN. I checked this against the diff and CI, not against the report's own account. Claim5853230223. It executes triage5852947380(bug · documentation · p2 ·area:devpath).ci-cd-pipeline.md×8,lookup.mdx×1 andselect.mdx×1. Commits, not PRs, because two landing PRs answer 404 too. Two prose-only mentions in the same sentences were re-pointed. No sentence's claim moved. +6/−6 across 3 fileslychee.toml: 10 × 404 on base, 0 objectui errors on head. The tree sweep finds 181 of 181 objectui links answering 200 acrosscontent/**,docs/**,README.mdandskills/**content/**belongs to no released packageCONTRACT_REVIEW_TIER: PASS at08be3d7, posted with this landing. The seat read the transcript: every model turn ran at tiergit merge-tree --write-treeagainst main25c7d584e4is clean. The three files are untouched on main since base08be3d7: 36 success, 3 skipped, 0 failure.check-linksruns onscheduleandworkflow_dispatchonly, so no PR shows it green. The triage asks for aworkflow_dispatchre-run after merge, and the seat's write channel has no dispatch op. The next weekly run (17 4 * * 0) is the instrument unless the maintainer dispatches itlabel-write --assignwas refused by the permission classifier; the seat has surfaced it to the maintainer rather than re-issue itFixes #10755onlyFindings, each with its disposition
lookup.mdxlines 68–76 saySelectOptionSchemarefusesdescriptionwith a 422. The installed spec 17.4.0 declares and accepts it, measured by the dev and by the review. The paragraph now cites its own landing commit, which dates the claim ⇒ objectui#10769.Generated by Claude Code
docs: cite the landing commits where ten objectui issue links answer …