check-doc-links 的包内 README 扫描只到顶层:4 个嵌套 README.md 仍未被任何门禁解析(实测 0 死链) #6026

Activity

1. yinlianghui-tw

   added

   finding

   on Aug 24, 2026
2. claude commented on Aug 25, 2026

   

   claudeBot

   on Aug 25, 2026 – with Claude

   Contributor

   More actions

   Concentrated triage batch: finding → pm:queue + domain:devx, Task, S — four nested README.md files are structurally outside both scan rows (basename exclude at any depth vs top-level-only globs); measured 0 dead links today, so this is coverage hygiene, not a live defect. Fix: narrow the #4938 row's exclude to top-level READMEs only (or add a nested-README row), keeping the no-double-parse and CHANGELOG exclusions intact; the four files in the body are the acceptance population.

   ---

   Generated by Claude Code
3. claude

   added

   domain:devxobjectui devx stream: fix lands on .github/, scripts/ or release pipeline — devx lane cross-repoobjectui devx stream: fix lands on .github/, scripts/ or release pipeline — devx lane cross-repo

   pm:queue

   and removed

   finding

   on Aug 25, 2026
4. claude

   added the

   Task

   issue type on Aug 25, 2026
5. yinlianghui-tw

   self-assigned this

   on Aug 25, 2026
6. yinlianghui-tw commented on Aug 25, 2026

   

   yinlianghui-tw

   on Aug 25, 2026 – with Claude

   ContributorAuthor

   More actions

   Claiming this issue.

   • Session: session_019b5UBNMtTzKbVtZZGvFuxe
   • Branch: claude/issue-6026-nested-readme-doc-links

   Will re-measure the four-file set against current main before implementing (the card warns packages/core/src/adapters/README.md was rewritten by #6227).

   ---

   Generated by Claude Code
7. yinlianghui-tw commented on Aug 25, 2026

   

   yinlianghui-tw

   on Aug 25, 2026 – with Claude

   ContributorAuthor

   More actions

   os-dev-report

   {
     "issue": 6026,
     "status": "done",
     "branch": "claude/issue-6026-nested-readme-doc-links",
     "pr": "https://github.com/objectstack-ai/objectui/pull/6279",
     "premise_still_valid": true,
     "summary": "The structural gap is real and the file set is UNCHANGED after re-measuring against origin/main @ ef2a3bd8d: the same four nested README.md files (components/src/__tests__, core/src/adapters, plugin-gantt/docs/verification, types/src/zod) are seen by no SCAN_ROOTS row. The card's LINK COUNTS are stale, though: it measured 10 decidable links in the gantt verification README and 0 in core/src/adapters; today they are 94 and 2, for a re-measured entry price of 4 files / 97 markdown links / 96 decidable / 0 dead (core/src/adapters/README.md was rewritten by #6227, as warned). Took option 1, with a twist: added a `collect` mode to walk() (the inverse of `exclude` - the only basenames kept, at every depth) plus two rows `packages/*/*` and `apps/*/*`. The SECOND wildcard segment is what keeps the top-level README out, so the no-double-parse guarantee is structural (the rows are rooted at each package's subdirectories, and the top-level README is not inside any of them) rather than a filter someone must keep in sync. Option 2 was rejected: it needs walk() to learn depth-relative-to-glob-root, which would then be load-bearing for every future row. CHANGELOG.md stays excluded at every depth, unchanged. The three package rows now partition each package directory, asserted repo-wide.",
     "tests": "All at final head 31e1e2557, exit codes captured by redirect before any pipe. GATE VERDICT LINES: (1) `pnpm run docs:check-links` exit 0 -> 'Links are valid across 17 scan roots.' (was 15 roots before). (2) `pnpm run check:control-bytes` exit 0 -> 'check-control-bytes: OK (scanned 5171 tracked text file(s); skipped 85 binary).' (3) `pnpm run type-check:scripts` exit 0, output echoes 'tsc -p tsconfig.scripts.json' as a positive control against a zero-match no-op; it was RED first with TS2345 (argument of type Set-of-string|undefined not assignable to parameter of type null|undefined) because the `collect = null` default made checkJs infer a null-only param - fixed with explicit @param JSDoc on walk() and collectFiles(). (4) `pnpm run lint:root` UNNARROWED exit 0 -> 'x 28 problems (0 errors, 28 warnings)', all warnings pre-existing (vitest.setup.base.ts etc), none in the changed files; population control via --format json = 196 files linted, with both changed files present in that list. (5) `pnpm exec vitest run scripts/__tests__/check-doc-links.test.ts --reporter=verbose` exit 0 -> 'Test Files 1 passed (1) / Tests 101 passed (101)', all 8 new tests collected BY NAME under the 'objectui#6026 - the nested READMEs' describe. (6) `pnpm exec vitest run scripts/__tests__` exit 0 -> 'Test Files 77 passed (77) / Tests 2215 passed (2215)'. BEFORE/AFTER SCAN COUNTS: scan roots 15 -> 17; files scanned 269 -> 273; markdown links scanned 1224 -> 1321; files opened by more than one row 0 -> 0; the four nested READMEs in the surface 0/4 -> 4/4; new per-row counts packages/*/* = 4, apps/*/* = 0. NON-VACUITY (the entry price is zero, so a green run proves nothing): repointed packages/core/src/adapters/README.md at ../../../data-objectstack/NOWHERE-6026.md; the mutation was confirmed ON DISK first (injected marker grep count 0 -> 1, old anchor count 2 -> 1, git diff --stat non-empty) - not inferred from the editor's exit code. NEW gate exit 1 -> 'Found 1 broken link (1 distinct target): - [example-relative] packages/core/src/adapters/README.md:127 -> ../../../data-objectstack/NOWHERE-6026.md'. The PRE-PR bytes of the gate (git show ef2a3bd8d:scripts/check-doc-links.mjs, run from a temp copy inside the worktree so repoRoot still resolves to this repo) on the IDENTICAL mutated tree: exit 0 -> 'Links are valid across 15 scan roots.' Reverted: git diff --exit-code returned 0, git status --porcelain empty, 0 residual occurrences of the marker anywhere in the tree, temp gate copy deleted. TEST ABLATION: with the two new SCAN_ROOTS rows deleted (row grep count 1 -> 0 confirmed on disk), the suite goes 'Tests 8 failed | 93 passed (101)', failing both real-tree surface assertions plus the two pinned-table tests; the ablation script carried trap '<restore>' EXIT INT TERM and the restore was verified byte-exact with git diff --exit-code (exit 0) and the row count back to 1. Note the partition test correctly stayed GREEN under that ablation - removing rows cannot create duplicates; it guards the opposite direction. DECLARED NARROWING: the full ROOT vitest suite was NOT run locally - the shared verify lock returned 'VERDICT queue-timeout (exit 99) - never acquired' twice for it (9m00s each, ~18 min total) while another agent held it for their own root run. Measurement justifying the narrowing: the diff is exactly two files, both under scripts/; a repo-wide grep over *.ts,*.tsx,*.mjs,*.js,*.cjs (node_modules excluded) finds exactly ONE importer of scripts/check-doc-links.mjs, namely scripts/__tests__/check-doc-links.test.ts; and every gate test that reads the scripts/ tree wholesale already lives in scripts/__tests__, which ran in full (77 files / 2215 tests). CHANGESET: none owed - the diff touches only scripts/, check-changeset-presence.mjs guards each versioned package's own src/**, and #6212 / #6216 / #6260 are same-shaped precedents that carried none. objectui has no skip-changeset label; an empty-frontmatter changeset is the declaration for guarded source that publishes nothing, which this is not.",
     "open_questions": [],
     "out_of_scope_findings": [
       "NOT FILED - GitHub search/list is rate-limited for this account right now (both search_issues and list_issues return 'API rate limit already exceeded for user ID 314596056'), so the mandatory pre-file duplicate scan could not be run and I did not file blind. Handing it to PM: TWO stale coverage claims about this very gate. (a) packages/components/src/__tests__/readme-shadcn-sync-categories.test.ts, in the '## Scan surface' docblock, says README_SHADCN_SYNC.md 'has never been scanned by anything' and that widening the scan root 'is a separate change with its own entry price'. Both became false with #4938: that file is not named exactly README.md, so the packages/* row already scans it today. The sentence is load-bearing - it is the stated justification for that test doing its own path checking. (b) content/docs/guide/ci-cd-pipeline.md carries the gate's surface list twice, at line ~406 ('content/docs/, examples/, the root README.md, CONTRIBUTING.md, ROADMAP.md, the internal docs/ tree and every package README.md') and in the two-link-checkers table at line ~972 (same list). Both stop at #3622 and omit #4148's app READMEs and root-level files and #4938's package/app directory trees - and now this PR's nested READMEs. Nothing gates either list. Deliberately NOT fixed in this PR: the staleness predates this change by two cards, and rewriting both surface lists is a docs change with its own review surface. Suggested labels: finding, domain:devx, no pm:queue."
     ]
   }

   ---

   Generated by Claude Code

   ---

   Generated by Claude Code
8. os-zhuang

   closed this as completedin #6279on Aug 25, 2026