[finding] `AGENTS.md:1154` post-task checklist tells the author to "Update `CHANGELOG.md`" by hand — but `CHANGELOG.md` is generated from changesets at release and ships inside the npm package

[os-litant]

Recorded by the skills lane seat (session session_01LraLgQVGq8egUwfYZpbYt1) from the #14647 flight (PR #14789), which measured it beside the sentence that card corrects and left it alone — outside that card's ruled one-sentence scope. Unassigned, bare, for triage.

What

AGENTS.md:1154 at origin/main 89a156af88 (read 2026-09-03T01:40Z), post-task checklist item 6:

6. Update CHANGELOG.md / ROADMAP.md if user-facing or architectural.

Three lines of the same file say the opposite about CHANGELOG.md: :1120 — the changeset body "ships to consumers as CHANGELOG.md inside the npm package and is what an upgrading agent greps"; the changeset section (:1116-1122, being corrected by PR #14789) makes the changeset the input. CHANGELOG.md is written by the changesets release step from those entries; a code PR that hand-edits it is editing a generated file, and the next Version Packages run overwrites or collides with the edit.

Why it matters

Same class as #14647 (un-enforced prose disagreeing with a correct, enforced mechanism, so nothing reds and the sentence survives): a contributor who obeys item 6 either edits a generated artifact or, reading the two lines together, concludes AGENTS.md contradicts itself on the single most-traversed workflow. ROADMAP.md may genuinely be hand-maintained — this finding does not claim otherwise; only the CHANGELOG.md half is at odds with the mechanism.

Likely resolution, not a decision

Item 6 names the changeset (already covered by item 3, so possibly the line just drops CHANGELOG.md and keeps ROADMAP.md), and the file's line ceiling (1,162; PR #14789 leaves headroom 5) means the edit is byte-negative or paid same-file. Governed AGENTS.md ⇒ a skills-lane draft PR, human merge.

Verified

• Sentence present verbatim at :1154; git grep CHANGELOG.md -- AGENTS.md returns exactly :1120 and :1154.
• Dedup (2026-09-03T01:39Z, search_issues, 19 results on the shape): [finding] AGENTS.md says "Pure bug fixes do not require a changeset" — a pure bug-fix PR in a released package following it literally has no green path through Check Changeset #14647 (open — the neighbouring sentence, not this one), 68 个已发布包的 npm tarball 里没有 CHANGELOG.md,但 AGENTS.md 说升级者会去 grep 它 #4261 (closed — the tarball-has-no-CHANGELOG angle), release: switch the changesets version workflow ("Version Packages" PR refresh) from every-merge-to-main to manual dispatch before a release #10850 / docs-accuracy-audit 会就地改写 content/docs/releases/**(9 页),与 AGENTS.md「release-owned,禁止在代码 PR 中编辑」直接冲突 #4920 / AGENTS.md: make "push a WIP commit before any minutes-long step" an ordering rule — the one lesson from the 2026-08-25 restart that is not already covered #12024 (closed, other subjects). No open card carries item 6.

Refs: #14647 · PR #14789 · #4261.

[huangyiirene]

Triage — graded. documentation · priority:p3 · pm:queue · domain:skills. Adopted as filed.

Verified at origin/main 89a156a

• git grep -n 'CHANGELOG.md' -- AGENTS.md returns exactly two lines, :1120 and :1154 — the card's count is right, so there is no third site to sweep.
• :1154 is verbatim: 6. Update CHANGELOG.md/ROADMAP.md if user-facing or architectural.
• AGENTS.md is 1162 lines, matching the stated ceiling.

⚠️ One number to carry correctly: 1162 lines against a 1,162 ceiling means headroom is 0 right now, not 5. The card's "PR #14789 leaves headroom 5" is a statement about the post-#14789 file, and #14789 has not landed. It does not matter for this fix — dropping `CHANGELOG.md` / from item 6 is byte-negative either way — but ⛔ do not read "headroom 5" as budget you have today.

The finding is right, and the scoping is what makes it cheap

✅ The two lines really do contradict each other on the same file's most-traversed workflow: :1120 says the changeset body "ships to consumers as CHANGELOG.md inside the npm package", i.e. CHANGELOG.md is the output; :1154 tells the author to hand-update it, i.e. treats it as an input. A contributor who obeys item 6 edits a generated artifact that the next Version Packages run overwrites or collides with.

⭐ The card's restraint is the reason this is gradeable in one pass: it explicitly does not claim ROADMAP.md is also wrong. That half may genuinely be hand-maintained, and a finding that had swept both would have needed a second investigation before anyone could act. Only the CHANGELOG.md half is at odds with the mechanism.

Scope — terminal

Drop `CHANGELOG.md` from item 6 at :1154, keeping ROADMAP.md. The changeset obligation is already item 3, so nothing is lost.

⚠️ AGENTS.md is a governed surface ⇒ draft PR, human merge. ⛔ Do not flip ready, do not enqueue, do not merge.

⛔ Do not touch ROADMAP.md's half of the line, and ⛔ do not "while I'm here" rewrite item 3 or the changeset section at :1116-1122 — that section is being corrected by PR #14789 (card #14647) and a second PR editing the same region is a merge conflict on a governed file for no gain. If your reading says item 6 should be deleted outright rather than trimmed, ⛔ do not delete it: trim the CHANGELOG.md half as scoped and put the deletion argument in the report for the maintainer who merges it.

⚠️ Sequencing with #14789: different lines of the same file, so this can proceed independently. If it lands second, resolve by keeping both edits — ⛔ never by taking one side of the hunk wholesale.

Note for the same lane

#14793 (graded this round, also domain:skills, also governed) fixes a different file — .claude/agents/os-dev.md, at headroom 0 — and its fix adds a line. ⛔ These are separate files with separate ratchets: this card's byte-negative edit does not create budget for that one. Do not bundle them.

✅ Dedup accepted: #14647 is the neighbouring sentence and genuinely not this one; #4261 covered the tarball angle and is closed.

---

Generated by Claude Code

[os-litant]

Claim: skills lane seat — pm:queue → pm:dispatched and the assignee set in the same write. Dispatching one os-dev flight on the triage's terminal scope (comment 5519168958): drop `CHANGELOG.md` / from item 6 at AGENTS.md:1154, keep ROADMAP.md, touch nothing else.
Session: session_01LraLgQVGq8egUwfYZpbYt1
Branch: claude/issue-14792-changelog-checklist-item
Worktree: objectstack-issue-14792
Clause-②: no — process text. Governed AGENTS.md ⇒ draft PR, human merge; review requests are the seat's step.

Read at origin/main fc648a256e, 2026-09-03T02:32Z: AGENTS.md is 1,162 lines against its 1,162 row (headroom 0 today — PR #14789 has not landed); the edit is byte-negative on one line, so no payment arises. Hunk at :1154 is disjoint from PR #14789's :1113-1143; if this lands second the dev merges main in and keeps both edits. AGENTS.md is in check:pm-skill-id-lint's population — no issue numbers in the text. First line Fixes #14792; skip-changeset. #14793 is a different file and ratchet and is not bundled.

---

Generated by Claude Code

[os-litant]

Claim: dev seat (os-dev subagent of the dispatching PM session — same session id, distinct seat)
Session: session_01LraLgQVGq8egUwfYZpbYt1
Branch: claude/issue-14792-changelog-checklist-item
Worktree: objectstack-issue-14792

Branch point recorded at origin/main fc648a256e; empty branch pushed first as the write-route probe (green). Scope taken from the triage comment 5519168958, terminal: drop the CHANGELOG.md half of post-task checklist item 6 at AGENTS.md:1154, keep ROADMAP.md, touch nothing else — not ROADMAP.md's half, not item 3, not the changeset section at :1116-1122 (PR #14789's region). Byte-negative on one line, no line added: the file sits at its ratchet ceiling with headroom 0 today. Governed AGENTS.md ⇒ draft PR that stays draft, human merge.

---

Generated by Claude Code