Repository navigation
scripts/adr-index.mjs reads header fields only from an ADR's first 40 lines, so a longer status note silently empties the Supersedes cell while check:adr-index still passes #30916
Description
repo:cloud#3 · hotlong · session_01Q4EjjRhUZjdy8VXs6Kkur2
Filing gate: (a) a measured defect in an internal tool, found by the objectstack#30881 dev and confirmed by contract review 6682453235 on PR #30899 (should-fix: "file a follow-up card").
维护者速读
生成 ADR 目录的脚本只读每份 ADR 的前 40 行。ADR 顶部的状态说明一旦变长，「Supersedes / Amends」等字段被挤到第 40 行之后，目录里那一栏就会悄悄变空，而目录检查照样通过。这次 ADR-0146 的修订刻意把说明压短，才没触发。下一次谁写长一点就会中招。
Measured
scripts/adr-index.mjs:HEADER_WINDOW = 40. It reads**Key**:header fields only within the first 40 lines.- ADR-0146 after the amendment:
Supersedesat line 38,Amendsat 39,Builds onat 40. Two lines of headroom remain. - An earlier draft of the amendment, with a 7-line status note, pushed
Supersedes in partto line 42. The regenerated index then emptied ADR-0146's Supersedes cell (ADR-0126, 0131, 0005, 0048, 0141 and 0070 became an em dash), and the run still exited 0. CI'scheck:adr-indexruns in--reportmode and always exits 0.
Expected
- Header parsing reads the whole header block (up to the first heading or the first blank-line-delimited body), not a fixed line window. Or it fails loudly when a known key appears after the window.
- The CI check fails when a regenerated index would drop a field that the previous index carried.
Acceptance
- A self-test with an ADR whose status note pushes
Supersedespast line 40: the index keeps the cell, or the tool exits non-zero.
Suggested-grade: p3 · an internal tool; no user door.
Activity
objectstack-fleet commentedon Nov 6, 2026
triage#1 · os-zhuang · session_01AavokzJ5DndAwitDXvKy4U
Path: none
Triage: audit · p3 ·
domain:devx· 2026-11-06T12:58Z⛔ Not a claim.
scripts/adr-index.mjsreads**Key**:header fields only from an ADR's first 40 lines (HEADER_WINDOW = 40). A longer status note silently empties the Supersedes cell, andcheck:adr-indexstill passes in--reportmode. Found on PR docs(adr): ADR-0146 amended in place — an upgrade plan may report a .custom override that hides a vendor change, as advice only #30899's contract review.domain:devxandpm:queuekept.Generated by Claude Code
objectstack-fleet commentedon Nov 7, 2026
domain:devx#3 · os-project-manager · session_01FEkQ5njkh3xbcDN6EtSfw7
Claim: PM loop round 1
Session:
session_01FEkQ5njkh3xbcDN6EtSfw7Account:
os-project-managerBranch:
claude/project-thread-timx6f(a project card thread,session_01FEkQ5of5Rharw5eu6Hvhes; the thread can push only its harness branch, a measured platform exception toclaude/issue-<n>-<slug>)Worktree:
objectstack-issue-30916Domain:
domain:devxSeat:
domain:devx#3File surface:
scripts/adr-index.mjsand its self-test, the rootpackage.jsoncheck:adr-indexline only if its invocation must change, and the generated ADR index only if regeneration changes it (it should not). ⛔ No rootci.ymledit.Container & model:
S,mode:thread,model: default — dispatch-gates --tier: no path-derived mandate,effort: mediumClause-②: no
Metadata-protocol: none
Responsibility: n/a — internal tooling defect, no user door
Thread-read: 6685367873
Premise: holds. On main
5b22ee759,scripts/adr-index.mjsstill declares a fixed header window (line 103:HEADER_WINDOW = 40), andcheck:adr-indexruns--report.Serial constraints cleared: none in flight on
scripts/adr-index.mjs; #30763 (in flight) touchesscripts/pm/dispatch-gates*only.Ruling-ref: 「p1 p2 处理完了，就可以处理 p3」 (maintainer, 2026-10-26T10:29Z) — the lane's unclaimed p1 cards are a decision card on the root CI workflow (#30461) and #31159, serialized behind PR #31204.
Order:
pm-orders/devx/order-30916.md(project files), handed to the card thread in its opening message.Generated by Claude Code
objectstack-fleet commentedon Nov 7, 2026
domain:devx#3 · os-project-manager · session_01FEkQ5of5Rharw5eu6Hvhes
os-dev-report
{
"issue": 30916,
"status": "done",
"branch": "claude/project-thread-timx6f",
"pr": "#31252",
"session": "session_01FEkQ5of5Rharw5eu6Hvhes",
"premise_still_valid": true,
"summary": "scripts/adr-index.mjs (+180/-22, one file): header fields are read from the whole header block (every line before the first heading after the title), not the first 40 lines; a supersession blockquote counts only when 'superseded' sits in a bold span, because the plain whole-block rule misreads ADR-0126 line 44 (prose about a draft of ADR-0131) into a Superseded-by cell, and ending the block at '---' instead loses ADR-0065 from ADR-0080/0081. All 160 records measured; regenerated docs/adr/README.md on 5b22ee7 is byte-identical. A Date/Supersedes/Superseded-by cell the committed index carried that the regeneration empties is a dropped field: --report exits 1 with an ::error annotation (every other finding stays a warning, exit 0), --check exits 1, the write mode writes and exits 1. The CI half needs no workflow edit: the lint ADR step is continue-on-error and Lane verdict reads its outcome, so a drop reds Lint & Repo Gates. Ablations done although os-dev.md lists scripts among the not-ablated kinds; the order's 'a new negative pin is ablated' line was followed as the per-card ask (one-time proof, nothing left in the tree). No claim of my own; card assignee not touched.",
"tests": "node scripts/adr-index.mjs --self-test on d8acb85: '✓ adr-index --self-test: 60 assertions' (was 46; new batteries header block 6, dropped fields 7; drift battery report pin rewritten; floor 7 -> 9). Ablations via scripts/ablation-replace.mjs, each anchor x1 -> x0 and blob moved, restored with git diff HEAD empty: window (headerBlock -> lines.slice(0, 40)) exit 1, 6 failures; notice (old blockquote regex) exit 1, a-plain-prose-mention-is-not-a-notice; report exit (return 0) exit 1, verdict-carries-the-drop-and-report-exits-1; no drop (if (false)) exit 1, 3 failures. Incident replay on copies of the real registry with 7 lines added to ADR-0146's status note (Supersedes in part line 39 -> 46): base script write exit 0 with ADR-0146 Supersedes cell an em dash; this branch keeps 'ADR-0126, ADR-0131, ADR-0005, ADR-0048, ADR-0141, ADR-0070', index unchanged. Drop path on the real tree (ADR-0123 Supersedes cell hand-set to ADR-0099): --report exit 1, --check exit 1, write exit 1, rerun exit 0 and README restored.",
"gates": "Derived by node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack on d8acb85: 31 commands, each run with exit captured before any pipe, all 31 exit 0; --ran: '31 derived famil(ies) accounted for — 31 run, 0 NOT-MEASURED'. Slowest: pnpm check:pm-dispatch-gates 449 s, check-comment-mask-corpus.mjs 205 s. pnpm check:adr-index: self-test 60 green, --report OK. Scoped lint: no package touched; pnpm exec eslint --no-inline-config scripts/adr-index.mjs exit 0. Not run locally (CI-only per dispatch-gates): Plan Test Core shards step, type-check lanes (no TypeScript in the diff). CI check-runs on #31252 not yet concluded at report time. Gate record: pm-orders/devx/30916-gates-ran.txt (project files).",
"dogfood_battery": "n/a — no
!in the PR title and no (narrowing) arm","mcp_calls": "4 — mcp__github__issue_read x3 (get, get_comments x2 on #30916), mcp__github__pull_request_read x1 (get #31252 read-back); no write tools",
"api_writes": "2 — POST /repos/objectstack-ai/objectstack/dispatches (relay pr_create: #31252, draft, labels skip-changeset, read back identical 9394/9394 bytes, assignees none); POST /repos/objectstack-ai/objectstack/dispatches (relay comment: this os-dev-report on #30916 via post-stamped.mjs). Plus one git push (not REST).",
"open_questions": [
{
"question": "lint.yml lines 688-696 (the comment above the 'ADR index is current' step, id adr-index-current-report-only) say 'So the only red this step can show is a broken generator, never a record'. After #31252 a dropped field also reds the step. The file is outside this claim's surface, so it was not edited. Who corrects the comment?",
"options": [
"A — the seat amends the claim to add lint.yml (comment lines only) and the fix rides #31252 as a second commit",
"B — leave #31252 as is and carry the one-comment correction as a follow-up docs-only change",
"C — no change: the script header, which the comment points at, is the authority and states the new behaviour"
],
"recommendation": "A, because the stale sentence is created by this PR and a comment-only edit adds no surface (startup axis), keeps the one statement of the step's behaviour true (long-term axis), and an AI reading the step comment would otherwise conclude a record can never red it (AI-error axis); real need: the step exists and its readers are agents. Not a new gate either way. This is a seat call, not a maintainer decision."
}
],
"security_findings": "withheld",
"out_of_scope_findings": []
}
Generated by Claude Code
objectstack-fleet commentedon Nov 7, 2026
domain:devx#3 · os-project-manager · session_01FEkQ5njkh3xbcDN6EtSfw7
Landed 2026-11-07T20:28Z · PR #31252 merged through the merge queue at 2026-11-07T20:26Z, as
bdefd49f30ed69bad7d9356a9ba66f8ad2ec9591.merged: true, merge commitbdefd49f30, head reviewedd8acb85f71(ACCEPT on the PR)completedbyFixes #30916State change:
pm:dispatchedremoved (card closed). Out of scope: #31261 filed (thelint.ymlADR index step's comment and name still say it never reds on a record; that file is governed rules text).