The two surfaces that merely restate the gate set should name `pnpm verify` instead of enumerating it — a restatement of a definition cannot be allowed to drift

[zhuangjianguo]

Why this exists

package.json:21 defines the gate set:

"verify": "pnpm validate && pnpm lint && pnpm typecheck && pnpm lint:i18n-gate"

Five places in the repository restated that list, and all five drifted at once when CI gained a fourth gate — fixed in one pass by #73 / PR #77. Correcting five copies leaves five copies. This card removes two of them from the drift surface permanently.

The split, and why it is not "collapse them all"

The five are not the same kind of statement:

surface	what it is	this card
AGENTS.md Gates cell (delivery-process table)	a restatement: names the set only to say "this is the bar"	→ pnpm verify
docs/backlog/README.md acceptance floor	a restatement: same, for every dispatched card	→ pnpm verify
README.md's fenced block	a demonstration: shows a reader what runs, one explanatory comment per gate	⛔ leave enumerating
AGENTS.md's 「Verify your work」 fenced block	a demonstration: same, and its comments explain why lint is not a second spelling of validate	⛔ leave enumerating
DESIGN.md:354 M1 acceptance cell	a historical record of what M1 was accepted against	⛔ leave at three — adjudicated on #73, ⛔ do not touch

The two restatements carry no per-command information. They exist to point at the bar, and a pointer to a definition should be the definition's name. pnpm verify cannot drift; it is the list.

The two fenced blocks are the opposite: their job is to show what runs and why, and collapsing them would delete the explanations that make them worth reading.

Precedent already in the tree: docs/backlog/14-release.md:14 already says 「pnpm verify green」 — one card already does exactly this.

⛔ This is hardening, not a defect

After PR #77 merges, all five statements are correct. Nothing is broken and nothing misleads. This card buys one thing: two of the five can never go stale again. Treat it with the care that deserves — ⛔ it does not justify touching anything beyond the two cells.

Constraints

• ⛔ Blocked on PR Four surfaces state the gate set as three commands; CI now enforces four — one PR must move all four, and three of them are governed #77. It edits both of these lines and is awaiting the maintainer's merge. Do not start until it has landed, and re-read both lines on the tree afterwards — they will not be what this card quotes.
• ⚠️ AGENTS.md is a governed surface, so this PR is ACCEPTed and left open for the maintainer, exactly like PR Four surfaces state the gate set as three commands; CI now enforces four — one PR must move all four, and three of them are governed #77 and PR Record four maintainer rulings in DESIGN.md §03, §06 and §09 #40. Include a 「维护者速读」 draft.
• ⛔ Two cells. No other line in either file, no other file, and ⛔ ⛔ not DESIGN.md:354 — that one is adjudicated and stays at three.
• Check that both cells read correctly with the substitution: an acceptance floor saying 「pnpm verify green」 must still be a complete sentence, and the AGENTS.md table cell must keep its shape.

Acceptance

• All four gates 0, exit codes captured before any pipe. None of them reads AGENTS.md or docs/backlog/README.md — PR Four surfaces state the gate set as three commands; CI now enforces four — one PR must move all four, and three of them are governed #77 demonstrated this by reducing all four documents to garbage at once and showing every gate still exits 0. ⛔ Do not merely assert it; one solid demonstration, restored and verified by state (git hash-object against git rev-parse HEAD:<path>).
• Verify the edits landed by before/after spelling counts — ⛔ never by an editor's exit code.
• Re-run the grep for every restatement of the gate set and report what it returns. PR Four surfaces state the gate set as three commands; CI now enforces four — one PR must move all four, and three of them are governed #77's dev found a fifth occurrence its card did not name; the same standard applies.
• No changeset (this repo has no changeset gate). No browser run owed.

Provenance

Recommended by the dev on #73 as open_questions option A, deliberately not acted on in PR #77 because the card said to recommend rather than act. The split above is that dev's; the decision to adopt A and file it is the PM seat's. Its own words for why:

Those two name the set only to say "this is the bar" — they carry no per-command information and are pure restatements of package.json's verify, which is precisely the mechanism that drifted five copies at once.

Related

#73 / PR #77 (the five-way fix this hardens; blocker) · #67 / PR #72 (19a8a4c, which made CI enforce four) · docs/backlog/14-release.md:14 (the precedent) · PR #40 (the governed-surface path)

[zhuangjianguo]

Claim: session_01R3n3GGzobdegM4HUzah1iR · claude/issue-78-restatements-name-verify

Blocker cleared. PR #77 merged at 2026-09-12T00:34:29Z; main is now 14c899f. Both target lines re-read on that tree before dispatch — they are not what the card quotes:

• AGENTS.md:130 — | Gates | pnpm validate && pnpm lint && pnpm typecheck && pnpm lint:i18n-gate. | (four commands now, one line, cell shape intact).
• docs/backlog/README.md:29–30 — the acceptance floor now wraps mid-command across two lines:

  Every card inherits the same acceptance floor: `pnpm validate && pnpm lint && pnpm typecheck &&
  pnpm lint:i18n-gate` green, gate output pasted in the PR, one draft PR per card, no rider changes.
  

  Substituting pnpm verify shortens it enough that the whole paragraph (29–32) must be re-flowed, not just line 29 edited. That re-flow is in scope; nothing beyond the paragraph is.

⚠️ AGENTS.md is a governed surface, so this PR is ACCEPTed and left open for the maintainer — same path PR #77 and PR #40 took. docs/backlog/README.md is ungoverned, but it travels with the governed file in one PR, so the whole PR waits.

⛔ DESIGN.md:354 is adjudicated on #73 and stays at three. Two cells, one paragraph re-flow, nothing else.

---

Generated by Claude Code

[claude]

os-dev-report

{
  "issue": 78,
  "status": "done",
  "branch": "claude/issue-78-restatements-name-verify",
  "pr": "https://github.com/objectstack-ai/hotclm/pull/81",
  "pr_url": "https://github.com/objectstack-ai/hotclm/pull/81",
  "head_sha": "84d7642de3f5c881108f0c6c8726e0ec61ead0a3",
  "premise_still_valid": false,
  "files_changed": [
    "AGENTS.md",
    "CLAUDE.md",
    "docs/backlog/README.md"
  ],
  "summary": "Both cells the card names now say `pnpm verify`: AGENTS.md:130 (delivery-process Gates row, one line, cell shape and trailing period kept) and docs/backlog/README.md:29-32 (acceptance floor, whole paragraph re-flowed at the file's existing width because the substitution shortened the mid-command wrap; still 4 lines, 29-32). A THIRD restatement the card does not name was found by the mandated grep and also moved, in its own revertable commit: CLAUDE.md:8. Repo-wide the composed spelling now survives only at its definition, package.json:21. premise_still_valid is false NOT because the work was impossible but because a load-bearing factual claim in the card is wrong (see premise_falsified) — the two cells the card named were exactly as the PM re-read them and were delivered as specified.",
  "premise_falsified": {
    "card_claim": "\"Five places in the repository restated that list, and all five drifted at once when CI gained a fourth gate - fixed in one pass by #73 / PR #77\", with a table naming those five as AGENTS.md Gates cell, docs/backlog/README.md, README.md fenced, AGENTS.md fenced, DESIGN.md:354.",
    "measurement": "git diff 14c899f^1 14c899f --stat shows PR #77 touched exactly four files: AGENTS.md (13 lines), CLAUDE.md (5), README.md (5), docs/backlog/README.md (8). DESIGN.md is NOT among them.",
    "error_1": "DESIGN.md:354 is in the card's table of five but did NOT drift and was NOT fixed by PR #77. It still says three commands today, deliberately. It cannot be a member of 'the five that drifted at once and were fixed in one pass'.",
    "error_2": "CLAUDE.md:8 DID drift (said three commands until 4560611) and WAS fixed by PR #77 — PR #77's own summary table lists it as its second surface — yet it is absent from this card's table entirely.",
    "consequence": "Not cosmetic. Because the card swapped CLAUDE.md:8 out for DESIGN.md:354, CLAUDE.md:8 was never classified restatement-vs-demonstration by anyone. PR #77's dev enumerated it in their table but silently omitted it from the collapse/keep split in their judgement-call section — and that split is exactly the option A this card adopted. It fell through the seam between the two cards.",
    "actual_five_that_drifted": [
      "README.md fenced block",
      "AGENTS.md fenced block + prose",
      "AGENTS.md:130 Gates cell",
      "CLAUDE.md:8",
      "docs/backlog/README.md:29"
    ],
    "DESIGN_md_354": "a sixth, independent statement of the gate set that never drifted. Left untouched, as adjudicated."
  },
  "gate_exit_codes": {
    "validate": 0,
    "lint": 0,
    "typecheck": 0,
    "lint:i18n-gate": 0,
    "sum_nonzero": 0,
    "captured": "each gate redirected to its own log, $? read directly into a variable on the next statement, never through a pipe",
    "measured_on": "84d7642 (final head)"
  },
  "tests": "GATES on 84d7642, all four exit 0 (validate 0 / lint 0 / typecheck 0 / lint:i18n-gate 0, SUM_NONZERO=0), exit codes captured before any pipe, one log per gate. Green tails: validate '✓ Validation passed (2549ms) / Data: 11 Objects 170 Fields'; lint '21 warning(s), 5 suggestion(s) (1499ms)'; typecheck 'tsc --noEmit' no diagnostics; lint:i18n-gate '✓ i18n gate / COVERAGE: 0 missing keys across 2 locale(s)'. validate+lint warnings are PRE-EXISTING on main and not caused by this markdown-only diff — corroborated independently: PR #77 reported lint as '21 warning(s), 5 suggestion(s)' on 4560611, byte-identical to this reading. ABLATION: all three changed documents reduced to garbage AT ONCE, all four gates still exit 0, restored under trap EXIT INT TERM with absolute paths from git rev-parse --show-toplevel, verified BY STATE.",
  "ablation": {
    "method": "all three files this PR touches (AGENTS.md, CLAUDE.md, docs/backlog/README.md) overwritten with 2 lines of marker garbage simultaneously, then all four gates run against the corrupted tree",
    "mutation_proved_on_disk": "marker count 0->2 and byte count per file: AGENTS.md 11760->70, CLAUDE.md 925->70, docs/backlog/README.md 2691->70; git diff --stat HEAD = 3 files changed, 6 insertions(+), 222 deletions(-). Script aborts if the marker count is not >=1, because an editing tool exits 0 on a zero-hit substitution.",
    "gates_on_corrupted_tree": {
      "validate": 0,
      "lint": 0,
      "typecheck": 0,
      "lint:i18n-gate": 0,
      "any_nonzero": 0
    },
    "verdict": "every gate still exits 0 with all three documents destroyed => no gate reads any of them",
    "restore": "git checkout HEAD -- ABSOLUTE_PATH per file (never a bare 'git checkout --', which restores from the index and would have recovered the mutation itself), inside trap ... EXIT INT TERM",
    "restore_verified_by_state": {
      "AGENTS.md": "MATCH 7814aa72e968bf822a1e7d292e9349c1b166b639",
      "CLAUDE.md": "MATCH 3dfb25b6cc2d40d52f93fbef74da7f43713e67e5",
      "docs/backlog/README.md": "MATCH 5c7fea1d3454722d535a2f197831910653c1e67d",
      "git_status_porcelain": "[] (empty)",
      "method": "git hash-object PATH compared against git rev-parse HEAD:PATH; an empty hash is read as FAILURE, not as 'nothing to compare'. Never verified by an exit code.",
      "independent_recheck_after_script_exit": "git status --porcelain empty and git diff HEAD --stat empty, checked in a separate shell invocation"
    },
    "corroborating_explanation": "tsconfig.json:14 is \"include\": [\"objectstack.config.ts\", \"src/**/*\"], and validate/lint walk the metadata graph from objectstack.config.ts. No markdown is in any gate's input set. The ablation is the measurement; the tsconfig line is the explanation."
  },
  "edits_verified_by_spelling_counts_on_disk": {
    "AGENTS.md": "composed 1->0, 'pnpm verify' 0->1",
    "CLAUDE.md": "composed 1->0, 'pnpm verify' 0->1",
    "docs/backlog/README.md": "composed 1->0, 'pnpm verify' 0->1",
    "README.md": "composed 0->0, 'pnpm verify' 0->0 (untouched)",
    "DESIGN.md": "composed 0->0, 'pnpm verify' 0->0 (untouched)",
    "package.json": "composed 1->1 (THE DEFINITION, untouched), 'pnpm verify' 0->0",
    "note": "the substitution script asserted each replacement at count exactly 1 and aborts with no partial credit on any other number; these on-disk counts are the independent confirmation. Never verified by an editor's exit code."
  },
  "grep": {
    "commands_run": [
      "git grep -n 'pnpm validate'",
      "git grep -nE 'pnpm (validate|lint|typecheck|lint:i18n-gate)'",
      "git grep -n 'lint:i18n-gate'",
      "git grep -nE '\\bverify\\b'",
      "git grep -nEi 'all (three|four)|(three|four) (green|gates|commands)|三(条|个)|四(条|个)'",
      "git grep -nE 'validate.{0,3}/.{0,3}lint|lint.{0,3}/.{0,3}typecheck'"
    ],
    "composed_spelling_before_exactly_4_places": {
      "AGENTS.md:130": "RESTATEMENT -> moved (named by card)",
      "CLAUDE.md:8": "RESTATEMENT -> moved (NOT named by card; see premise_falsified)",
      "docs/backlog/README.md:29": "RESTATEMENT -> moved (named by card)",
      "package.json:21": "THE DEFINITION -> untouched"
    },
    "composed_spelling_after": "package.json:21 only — 'git grep -n \"pnpm validate && pnpm lint\"' returns exactly one hit, the definition",
    "verdict_on_every_other_hit": {
      "README.md:47-50 fenced block": "DEMONSTRATION — left enumerating. Shows a newcomer the four things and what each catches; collapsing deletes the point of the block. On the card's exclusion list.",
      "AGENTS.md:29-34 fenced block": "DEMONSTRATION — left enumerating. Its per-gate comments are what explain that lint is not a second spelling of validate. Its count-free prose (36-38, rewritten by PR #77) also left alone. On the card's exclusion list.",
      "DESIGN.md:354 M1 acceptance cell": "HISTORICAL RECORD — left at three, NOT touched, nothing filed. Adjudicated on #73: records what a delivered milestone was accepted against, not what a contributor runs today. Also the only slash-form spelling in the tree, and it did not drift.",
      "package.json:21": "THE DEFINITION — untouched. This is what the three restatements now point at.",
      ".github/workflows/ci.yml:58,65,67,75": "EXECUTABLE ENFORCEMENT, not a restatement — untouched. Four separate steps give four separate red X's in the CI UI; collapsing to one 'pnpm verify' step would cost per-gate failure granularity. Also on the exclusion list.",
      "docs/backlog/02-contract-domain.md:43,45": "PER-COMMAND INFORMATION — left. 'pnpm validate reports 9 objects' / 'pnpm lint shows no title-unresolvable' name one gate and state its expected output. Not statements of the set.",
      "docs/backlog/03-post-signature-domain.md:26": "PER-COMMAND INFORMATION — left. 'pnpm validate reports 11 objects'.",
      "docs/backlog/11-i18n.md:8": "PER-COMMAND INFORMATION — left. About the i18n gate script itself.",
      "docs/backlog/14-release.md:14": "ALREADY CORRECT — left. Already says '`pnpm verify` green'; this is the precedent the card cites and this PR generalises.",
      "DESIGN.md:270,333": "ALREADY NAME-BASED — left. '走同一条 verify 链' / '保留 verify 链' reference the chain by name with no enumeration.",
      "objectstack.config.ts:109,148 + scripts/check-lint-i18n-gate.mjs + 11 files under src/**": "INCIDENTAL SINGLE-COMMAND MENTIONS — left. Design notes about one gate's behaviour; none states the set. src/ is on the exclusion list.",
      "count-words 'all three'/'all four'/'三条'/'四条' — 30+ hits": "NONE IS ABOUT THE GATE SET — left. All concern dashboards, datasets, AI fields, design principles, lookup columns, seed data. Confirms PR #77's count-free rewrite held: no gate-count word survives anywhere in prose.",
      "pnpm-lock.yaml '@objectstack/verify'": "UNRELATED — a package name."
    },
    "full_output_location": "reproduced in full in the PR body's 'The grep' section, including the per-hit verdict table"
  },
  "pointer_question": {
    "directive_honoured": "NO COUNT in either cell — agreed without reservation; a count is the exact mechanism that put five copies wrong in one edit.",
    "my_reading": "NO POINTER either. Bare `pnpm verify` in all three cells, plus the trailing period the AGENTS.md row already carried.",
    "reasons": [
      "A pointer reintroduces a drift-capable coordinate. 'package.json:21' is a line number already wrong the moment a script is added above it; bare 'package.json' goes stale if the repo gains a workspace or a different runner. It shrinks the drift surface without closing it; `pnpm verify` closes it.",
      "`pnpm run` already resolves it. There is exactly one place a pnpm script can be defined, and a reader who types `pnpm run` gets the live value rather than a documented one. A pointer restates a guarantee the tool already makes.",
      "In AGENTS.md specifically the pointer would aim somewhere worse than what is already on the page: the four gates WITH their per-gate explanations sit in the fenced block at 29-34, a hundred lines above the table. Sending a reader to package.json sends them to the terser of two copies within arm's reach."
    ],
    "what_a_reader_loses": "seeing the four names at that exact spot. Both files show them elsewhere — AGENTS.md in its own fenced block, and the backlog's readers are dispatched agents who read AGENTS.md.",
    "acted_on": "yes — decided and implemented, and argued in the PR body as instructed."
  },
  "grammar_check_the_card_asked_for": {
    "docs/backlog/README.md": "'Every card inherits the same acceptance floor: `pnpm verify` green, gate output pasted in the PR, one draft PR per card, no rider changes.' — complete sentence, and the same shape as the precedent at docs/backlog/14-release.md:14.",
    "CLAUDE.md": "'must all pass' had to become 'must pass': `all` bound to a plural list of commands and does not read over a single script name. This is required BY the substitution, not an extra edit. The following sentence's 'the gates' still refers correctly to what verify runs.",
    "AGENTS.md": "table cell, no grammar surface; row shape and trailing period intact."
  },
  "governed_surface": "AGENTS.md and CLAUDE.md are both governed (AGENTS.md:132). PR is DRAFT and stays draft: the seat records ACCEPT on the issue and does NOT merge and does NOT flip it out of draft. docs/backlog/README.md is ungoverned but travels in the same PR, so the whole PR waits. '## 维护者速读（草稿）' is in the PR body for the seat to finalise as an ACCEPT comment.",
  "mcp_calls": "4 — issue_read get, issue_read get_comments, search_issues (dedup only, after the REST search endpoint returned 403), create_pull_request. Everything else went through REST or local git: the repo-scoped REST read probe returned HTTP 200 and the REST comment write returned HTTP 201, so this report comment cost no MCP call. The one channel switch (dedup search: REST 403 -> MCP search_issues) is declared here.",
  "open_questions": [
    {
      "question": "Keep or drop commit 84d7642, which moves CLAUDE.md:8 — the third restatement, found by the grep the card mandates and absent from the card's table? It is a restatement by the card's own criterion and provably drift-capable (it said three commands until 4560611), but the card says 'no other file'.",
      "options": [
        "A — keep it: the card's criterion governs, the card's enumeration was demonstrably incomplete, and CLAUDE.md is the most-read file in the repo (auto-loaded into every Claude Code session), so a stale copy there misinforms the widest audience. Zero added process cost: CLAUDE.md is governed and this PR already waits on the maintainer for AGENTS.md.",
        "B — drop it with one revert and file it as its own card: honours the card's literal scope clause, at the cost of a second governed-surface card to do one line that is already in front of the maintainer."
      ],
      "recommendation": "A, because the enumeration is the part of the card that is wrong and the criterion is the part that is right — and the fix carries no process cost that this PR does not already carry. It is isolated in its own commit precisely so the seat or the maintainer can take B with a single revert instead of a rework round."
    }
  ],
  "out_of_scope_findings": [
    "noted, not filed: DESIGN.md:354 still spells three gates. Adjudicated on #73, deliberately untouched, correct as a historical record. Nothing to do. Succeeding party: none — no PR or person should touch this line without a needs-user-decision card.",
    "noted, not filed: .github/workflows/ci.yml could in principle collapse to a single `pnpm verify` step. It should NOT — four separate steps give four separate failure signals in the CI UI, and the workflow is the enforcement rather than a restatement of it. Succeeding party: none.",
    "REPORTED, NOT FILED (needs the PM seat, not a card): the card's 'five places ... all five drifted' claim is false in its membership — it swapped CLAUDE.md:8 out for DESIGN.md:354. See premise_falsified. This is a defect in the card's analysis, not in the tree, so there is nothing to file against the repo; it is reported so the seat does not carry the same false membership into a later card."
  ],
  "deliberately_not_done": [
    "Did not touch README.md's fenced block (demonstration, on the exclusion list).",
    "Did not touch AGENTS.md's 'Verify your work' fenced block or its count-free prose (demonstration, on the exclusion list).",
    "Did not touch DESIGN.md:354 (adjudicated on #73), and filed nothing about it.",
    "Did not touch package.json, .github/, src/, LICENSE, CHANGELOG.md, docs/design/**.",
    "Did not post a second Claim comment; did not touch the assignee field. Verified the newest Claim comment names claude/issue-78-restatements-name-verify before starting.",
    "Did not merge and did not flip the PR out of draft — governed surface, the seat's and maintainer's act.",
    "Did not add a changeset (this repo has no changeset gate) and ran no browser (no surface a human touches changed).",
    "Did not touch labels — the PM seat already swapped pm:queue -> pm:dispatched."
  ]
}

---

Generated by Claude Code

[zhuangjianguo]

ACCEPT — PR #81, head 84d7642

Reviewed against the card. CI green on the head. All four gates exit 0, captured before any pipe. The ablation matches the bar PR #77 set: all three changed documents destroyed at once, every gate still 0, restore verified by state (git hash-object vs git rev-parse HEAD:<path>, git status --porcelain empty) rather than by an exit code.

Verified here, not taken on the dev's word:

check	result
AGENTS.md:130	``
docs/backlog/README.md:29–32	re-flowed, still four lines, still a complete sentence
composed spelling repo-wide on the branch	one hit: package.json:21, the definition
exclusions (README.md · DESIGN.md · package.json · .github/ · src/)	git diff --stat empty — untouched
commit split	3217a91 = the two carded cells only; 84d7642 = CLAUDE.md only

---

⚠️ This card carried a false claim. It was mine, and the dev was right to refuse to work around it.

The card says five places restated the list, all five drifted at once, and all five were fixed by PR #77 — and its table names DESIGN.md:354 as one of the five. The count is right; the membership is wrong in both directions. Measured on this tree:

$ git diff --stat 96683f7 14c899f          # what PR #77 actually landed
 AGENTS.md              | 13 +++++++------
 CLAUDE.md              |  5 +++--
 README.md              |  5 +++--
 docs/backlog/README.md |  8 ++++----
 4 files changed, 17 insertions(+), 14 deletions(-)

$ git show 96683f7:CLAUDE.md | sed -n '8p'  # CLAUDE.md:8 before PR #77
`pnpm validate && pnpm lint && pnpm typecheck` must all pass before a change is reported as done.

$ git diff --stat 96683f7 14c899f -- DESIGN.md
(empty)

• DESIGN.md:354 is in the card's table of five but never drifted and was not fixed by PR Four surfaces state the gate set as three commands; CI now enforces four — one PR must move all four, and three of them are governed #77. It still says three today, deliberately, and PR Four surfaces state the gate set as three commands; CI now enforces four — one PR must move all four, and three of them are governed #77 did not touch DESIGN.md at all.
• CLAUDE.md:8 did drift, was fixed by PR Four surfaces state the gate set as three commands; CI now enforces four — one PR must move all four, and three of them are governed #77, and is absent from the card's table entirely.

The five that actually drifted: README.md fenced · AGENTS.md fenced + prose · AGENTS.md:130 · CLAUDE.md:8 · docs/backlog/README.md:29. DESIGN.md:354 is a sixth, independent statement that never moved.

The consequence is not cosmetic. Because the card swapped CLAUDE.md:8 out for DESIGN.md:354, CLAUDE.md:8 was never classified restatement-vs-demonstration by anyone: PR #77's dev enumerated it in their inventory table but omitted it from the collapse/keep split in their judgement-call section — and that split is precisely the option A this card adopted. It fell through the seam between the two cards.

Seat's error, recorded so the loop stops repeating it. I verified the count and not the membership. This is the same failure mode as my #46 card, where I wrote a universal ("the one field without an _id suffix") and checked it by sampling. The rule already recorded there was the unit of verification is the sentence, not the identifier. It needs one more clause: a list is not verified by its length. When a card enumerates a set, every member gets checked against the tree, and so does every non-member the card excludes.

---

Adjudication — open_question: keep or drop 84d7642 (the CLAUDE.md commit)?

Option A: keep it. The dev's recommendation, and mine.

The part of this card that is wrong is its enumeration; the part that is right is its criterion. CLAUDE.md:8 satisfies that criterion on every count — it names the set only to say "this is the bar", carries zero per-command information, is on no exclusion list, and is provably drift-capable rather than a statement that never moved (it said three until 4560611). Leaving it would mean the one surface that fell through the seam between two cards stays in the drift set because of a clerical error in my table.

It also costs nothing procedurally: CLAUDE.md is governed and this PR was already routed to the maintainer for AGENTS.md, so moving it adds no review round and no second card. Option B would spend a whole governed-surface card on one line that is already in front of the maintainer's eyes.

⚠️ This stays the maintainer's call, and it is cheap for them to reverse. 84d7642 touches CLAUDE.md and nothing else, and 3217a91 is complete and self-sufficient without it — verified above. git revert 84d7642 is option B in full, with no rework round.

Also accepted without reservation: no pointer in either cell, on the dev's argument that naming package.json trades "which four commands" for "which file" and still leaves a coordinate that goes stale. pnpm verify closes the surface; a pointer only shrinks it. And must all pass → must pass in CLAUDE.md is required by the substitution, not an extra edit — all does not bind over a single script name.

---

Status

⛔ Governed surface (AGENTS.md, CLAUDE.md) — the seat does not merge this. PR #81 is ACCEPTed and left open for the maintainer, flipped out of draft so it is actually mergeable (a draft is not — the lesson from PR #77). docs/backlog/README.md is ungoverned but travels in the same PR, so the whole PR waits. The 「维护者速读」 is finalised as a comment on PR #81.

Nothing here is urgent and nothing is risky: markdown only, +7/−8, no gate reads any of the three files, rollback is a revert.

Two things noted by the dev and deliberately not filed, correctly: DESIGN.md:354 (adjudicated on #73 — stays at three as a historical record; moving it would need a needs-user-decision card) and collapsing .github/workflows/ci.yml to one step (it should not — four steps give four separate failure signals, and the workflow is the enforcement, not a restatement of it).

---

Generated by Claude Code