Published packages/spec/CHANGELOG.md still states the patch-only undoable capture set, and contradicts itself four lines later

Path: none | 已发布面说错一句(北极星「优先级」4) | undoable 的捕获集在同一条 CHANGELOG 条目里前后矛盾,且已随 17.4.0 发布
分诊重测与定级:2026-09-20T15:25Z

What was measured

packages/spec is published (private unset, 17.4.0) and its files[] ships CHANGELOG.md. Read first-hand by this seat at origin/main = adf4b18777d507236cd24b7ed59b45a7c71bd1fd, 2026-09-20T10:31Z:

packages/spec/CHANGELOG.md, inside the single entry for effae80: feat(spec): a row action gets the declarative single-record field write (#14092):

:3918  - The action's existing `params`, `visible`, `confirmText` and `undoable` keys are reused,
         nothing duplicated; `undoable` now has its anchor — the patch names exactly the fields
         whose prior values are captured.

:3922  … `undoable` captures the prior values of exactly the fields written.

The same changelog entry states both sets, four lines apart. This is the same contradiction card #19148 found in packages/spec/src/ui/action.zod.ts (:1116 written set vs :1313 patch-only); PR #19283 settled the source file on the written set and corrected the describe text and the three generated reference pages, but a code PR may not edit content/docs/releases/ or a release-owned changelog, so this copy survives at head.

Why it is a card and not a fix in #19283

Review checklist: 「本轮令其变假或触碰的已发布缺陷必修;只注意到、未触碰的立卡不修,记明已发布面」. #19283 did not touch this file. The published surface is named above: packages/spec/CHANGELOG.md, shipped by files[].

For the record, #19283's PR body says the old sentence is "all corrected or regenerated here". That is overstated by this one line — noted so nobody reads it as a clean sweep. The PR is otherwise correct and the at-tier review PASSed it (comment 5749186061).

Direction is not settled here

A changelog is a historical record of what a release shipped, so "correct the line" and "leave it, it is history" are both defensible, and the entry's own :3922 already carries the accurate sentence. This seat chooses neither. What is measured is only that a published file states a promise the contract no longer makes, and that it contradicts itself in place.

Dedup words

CHANGELOG, undoable, prior values are captured, effae80, release-owned prose

Origin: at-tier contract review of PR #19283, comment 5749186061 (2026-09-20T10:18Z), "Old-sentence survivors at head, for the seat". Re-measured by this seat at 2026-09-20T10:31Z.
Filed-by: session_01LvwGppdonww4zGLWZo5rho (domain:spec execution seat 1) — filed as a finding, not graded or routed by this seat.

Generated by Claude Code

Claim: PM loop round 1
Session: session_013RDBh5DqXd2xnLwvHLgLFr
Branch: claude/issue-19301-changelog-undoable-capture-set
Worktree: objectstack-issue-19301
Domain: domain:spec
Seat: domain:spec#1
File surface: packages/spec/CHANGELOG.md — the single effae80 entry only (stop on breach; explain in the report)
Container & model: S, mode:subagent, model: claude-opus-5 — default judgment tier, quoting this act's own --tier run: 「no path-derived mandate … floor sonnet · default opus · ceiling opus」. ⛔ Not dropped to the floor: the card's own text says the DIRECTION is unsettled, so this is judgment work, not mechanical text replacement.
Clause-②: no
Thread-read: none
Serial constraints cleared: packages/spec/CHANGELOG.md is declared by none of the 13 in-flight claims in this lane, all read first-hand and intersected locally at 2026-09-22T04:57Z. git ls-remote shows no claude/issue-19301* branch ⇒ no pushed work, no implicit claim.

⏱️ Premise re-verified by this seat on origin/main 1912237481 at 2026-09-22T04:57Z, ⛔ not carried from the card. Both sentences are live in the one effae80 entry and they do contradict: :3918 「the patch names exactly the fields whose prior values are captured」 against :3922 「undoable captures the prior values of exactly the fields written」. Action surface: one recent commit on the file (c9b23cd, #19480, unrelated concern). Cited card #19283 is closed and settled the source file on the written set ⇒ :3918 is the stale half and :3922 already reads correctly.

Generated by Claude Code

os-dev-report

{
  "issue": 19301,
  "status": "done",
  "branch": "claude/issue-19301-changelog-undoable-capture-set",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/19661",
  "session": "session_013RDBh5DqXd2xnLwvHLgLFr",
  "premise_still_valid": true,
  "summary": "Re-verified first-hand in my own worktree at origin/main 5ce3705 (the seat's 1912237481 had already moved): the single effae80 entry in packages/spec/CHANGELOG.md did state both capture sets, found by text rather than by line number. Corrected the stale half in place so it names the merged write bag (`patch` UNDER the collected `params`), leaving the executor-contract sentence untouched — one line changed, no other entry, no restyling. Added the measured `patch` changeset. Draft PR #19661 is docs-only, which is the exact route AGENTS.md prescribes for a factual error in a released entry.",
  "tests": "Gate families derived in-worktree (`node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack --commands`, no paths passed) at 3c10091 = 61 families; all 61 run with $? captured to a file before any pipe; `--ran` reconciliation with exit codes recorded reports '61 derived, 57 run, 4 NOT-MEASURED, 0 UNRUN'. Through the shared verify lock (OS_VERIFY_LOCK_SLOT=issue-19301-spec-test, waited 0s each): `pnpm --filter @objectstack/spec test` VERDICT command-exit 0 — 511 files passed | 1 skipped, 14951 tests passed; `pnpm --filter @objectstack/spec typecheck` VERDICT command-exit 0; `pnpm --filter @objectstack/spec exec vitest run src/system/compliance-families-retirement.test.ts` VERDICT command-exit 0 — 10 tests (that suite walks CHANGELOG.md files and skips this filename by name). `pnpm lint` narrowing MEASURED, three pieces: (1) population read from eslint.config.mjs — every `files:` block targets {ts,tsx,mts,cts,js,jsx,mjs,cjs}, no markdown processor configured; (2) `eslint --no-inline-config --format json` over both changed paths returns 2 file results, each carrying only 'File ignored because no matching configuration was supplied.', errorCount 0 / rule warnings 0; (3) invariance — the diff holds no file in eslint's population and moves no TS/JS byte, so no untouched file's verdict can move. No ablation or reverse verification: the deliverable is a prose correction with no guard to ablate; the 'which set shipped' question was settled by reading the executor and its existing pin test instead.",
  "gates": {
    "node scripts/check-adr-0087-registration.mjs --base origin/main": 0,
    "node scripts/check-adr-0087-registration.mjs --self-test": 0,
    "node scripts/check-changeset-no-major.mjs --base origin/main": 0,
    "node scripts/check-changeset-no-major.mjs --self-test": 0,
    "node scripts/check-ci-filter-parity.mjs": 0,
    "node scripts/check-closing-keyword-parity.mjs": 0,
    "node scripts/check-closing-keyword-parity.mjs --self-test": 0,
    "node scripts/check-comment-mask-adoption.mjs": 0,
    "node scripts/check-comment-mask-adoption.mjs --self-test": 0,
    "node scripts/check-comment-mask-corpus.mjs": 0,
    "node scripts/check-dev-prereqs.mjs --self-test": 0,
    "node scripts/check-empty-changeset.mjs --base origin/main": 0,
    "node scripts/check-empty-changeset.mjs --self-test": 0,
    "node scripts/check-keyed-text-bounds.mjs": 0,
    "node scripts/check-keyed-text-bounds.mjs --self-test": 0,
    "node scripts/check-platform-object-tenancy-census.mjs": 0,
    "node scripts/check-platform-object-tenancy-census.mjs --self-test": 0,
    "node scripts/check-plugin-teardown-shape.mjs": 0,
    "node scripts/check-plugin-teardown-shape.mjs --self-test": 0,
    "node scripts/check-registry-log-declared.mjs": 0,
    "node scripts/check-registry-log-declared.mjs --self-test": 0,
    "node scripts/check-release-section-coverage.mjs": 0,
    "node scripts/check-release-section-coverage.mjs --self-test": 0,
    "node scripts/check-release-section-coverage.mjs --strict": 0,
    "node scripts/check-rest-log-spy-declared.mjs": 0,
    "node scripts/check-rest-log-spy-declared.mjs --self-test": 0,
    "node scripts/check-system-context-census.mjs": 0,
    "node scripts/check-system-context-census.mjs --self-test": 0,
    "node scripts/check-undeclared-dep-imports.mjs": 0,
    "node scripts/check-undeclared-dep-imports.mjs --self-test": 0,
    "node scripts/docs-audit/check-affected-docs.mjs": 0,
    "node scripts/docs-audit/check-drift-comment.mjs": 0,
    "node scripts/pm/release-rehearsal-clone.mjs --self-test": 0,
    "pnpm --filter @objectstack/spec run check:empty-state": 0,
    "pnpm --filter @objectstack/spec run check:liveness": 0,
    "pnpm --filter @objectstack/spec run check:strictness-ledger": 0,
    "pnpm --filter @objectstack/spec run check:variant-docs": 0,
    "pnpm check:changeset-gate-self-tests": 0,
    "pnpm check:cross-package-test-inputs": 0,
    "pnpm check:doc-authoring": 0,
    "pnpm check:driver-memory-census": 0,
    "pnpm check:dts-closure": "NOT MEASURED (exit 3 — PREREQUISITE NOT MET: no package has a dist/ in this worktree)",
    "pnpm check:dual-build-cjs-loads": "NOT MEASURED (exit 3 — PREREQUISITE NOT MET: no package has a dist/ in this worktree)",
    "pnpm check:gitlink-declared": 0,
    "pnpm check:issue-citations": 0,
    "pnpm check:lean-entry-closure": "NOT MEASURED (exit 3 — PREREQUISITE NOT MET: no package has a dist/ in this worktree)",
    "pnpm check:merge-driver": 0,
    "pnpm check:nul-bytes": 0,
    "pnpm check:objectui-changeset": 0,
    "pnpm check:org-identifier": 0,
    "pnpm check:page-declaration-shape": 0,
    "pnpm check:pm-changeset-deadline-census": 0,
    "pnpm check:published-files": 0,
    "pnpm check:refd-timer-probe": 0,
    "pnpm check:release-notes": 0,
    "pnpm check:release-page-status": 0,
    "pnpm check:slot-lookup": 0,
    "pnpm check:sourcemap-no-sources-content": "NOT MEASURED (exit 3 — PREREQUISITE NOT MET: no package has a dist/ in this worktree)",
    "pnpm check:test-source-alias": 0,
    "pnpm check:tier-file-adoption": 0,
    "pnpm check:watch-hint-literal": 0
  },
  "files_changed": [
    "packages/spec/CHANGELOG.md (1 line, inside the single effae80 entry)",
    ".changeset/19301-changelog-undoable-capture-set.md (new, patch)"
  ],
  "deviations": [
    "DECLARED FILE-SURFACE ADDITION: the dispatch declared `packages/spec/CHANGELOG.md` (effae80 entry only). A second path was added — `.changeset/19301-changelog-undoable-capture-set.md` — because the changeset question you asked me to measure came out `patch`, and the `Check Changeset` gate (pr-automation.yml) is path-agnostic: any PR adding no changeset and carrying no `skip-changeset` label goes red. The CHANGELOG surface itself was not exceeded: exactly one line inside the one entry moved.",
    "NARROWING (declared): 4 of the 61 derived families are NOT MEASURED — check:dts-closure, check:dual-build-cjs-loads, check:lean-entry-closure, check:sourcemap-no-sources-content, each exit 3 'PREREQUISITE NOT MET' because no package carries a dist/ here. They read built output across the whole workspace; a markdown-only diff cannot reach a dist/. Not run: a full `pnpm build` to satisfy them would hold the shared lock for a whole-farm build with no reachable effect. CI Build Core covers them.",
    "NOT MEASURED (with reason): the two cross-package `action-owner-key-single-source.test.ts` suites (packages/runtime, packages/objectql) exit 1 on 'Failed to resolve entry for package @objectstack/objectql' in an unbuilt worktree — module resolution, not a finding. Read instead: their SCANNED_EXTENSION is '.ts' and both `CHANGELOG.md` and `.changeset/` are ruled exclusions in their NOT_A_STALE_MENTION table, so this diff is invisible to them by construction.",
    "ZERO LABEL WRITES: the dispatch named no label, and `skip-changeset` is measured as not applicable, so the label set is empty — one budgeted endpoint (POST /issues/{pr}/labels) deliberately unspent.",
    "INSTRUMENT REJECTED, not cited: `turbo ls --affected` reports 80 of 81 packages in this worktree (76 with TURBO_SCM_BASE=origin/main), i.e. no usable narrowing signal here. I did not cite it as evidence rather than quote a reading I cannot defend.",
    "Commit trailers use the model-free pair per AGENTS.md, not the harness reminder's `Co-Authored-By: Claude Opus 5` form; the pre-push hook confirmed 'no card relation and no model identifier in the trailer pair'."
  ],
  "premise_corrections": [
    "The entry is under the `## 17.3.0` heading, NOT 17.4.0. Your Zone 1 wording ('what 17.4.0 actually shipped') and the card's '已随 17.4.0 发布' are both defensible about the TARBALL (CHANGELOG.md accumulates and ships in the current 17.4.0 package) but not about the RELEASE the entry documents. The PR body states 17.3.0 explicitly.",
    "Zone 2 item 1 held: both sentences live, found by text. Zone 2 item 2 held and is now evidenced: `declarativeUpdateWrite` returns `{ ...patch, ...params }`. Zone 2 item 3 held: #19283 is a closed PR, #19148 a closed issue. Zone 2 item 4 held: c9b23cd is the one recent commit on the file.",
    "STRONGER than your Zone 1 reasoning: the written set is not merely the set #19283 settled on — it is the set that SHIPPED. `executeDeclarativeUpdateAction` keys undoData off Object.keys of the merged bag, and packages/runtime/src/action-declarative-update.test.ts already pins '`undoable: true` carries the prior values of EXACTLY the fields written'. So the corrected sentence was wrong the day it was written; this is an amendment toward what shipped, not a retrofit of a later decision. The 'leave it, it is history' reading would preserve a sentence that never described any release.",
    "The route question you asked me to check: AGENTS.md Documentation Guardrails does constrain this path — `packages/*/CHANGELOG.md` is RELEASE-OWNED, 'Never edit in a code PR' — but it PRESCRIBES this delivery in the same cell: 'Factual error in a released entry → amend that entry in a dedicated docs-only PR, ⛔ never an erratum in a later entry and never a rider on code changes'. So: no STOP, and the shape is forced — docs-only, amendment in place, no code riding along. packages/spec/CHANGELOG.md is on no governed-surface row (Prime Directive 14), so no tier record is owed."
  ],
  "changeset_argument": "patch, NOT skip-changeset. Measured, three ways: (1) packages/spec/package.json — `private` unset, version 17.4.0, and `files[]` literally names CHANGELOG.md; (2) empirical — `npm pack --dry-run --ignore-scripts --json` in packages/spec lists 277 files with CHANGELOG.md among them, so the bytes this PR changes ship in the tarball; (3) the rule — AGENTS.md Post-Task Checklist step 3 says skip-changeset 'is for a diff that publishes nothing from any released package', which this is not. Repo precedent agrees: .changeset/18510-… (a JSDoc sentence) and .changeset/15295-… (a code comment) are both `patch`, as is .changeset/19148-undoable-capture-set.md, the still-pending changeset for the source-file half of this same sentence. Note the one argument AGAINST that I considered and rejected: a changeset lands a note in a LATER release entry, and AGENTS.md forbids 'an erratum in a later entry'. That prohibition is about where the CORRECTION lives — this correction is an amendment in place, so the reader who greps the old promise lands on the corrected entry; the changeset is the release note for the fix, not a stand-in for it.",
  "mcp_calls": "0 — no MCP tool of any kind was called; every GitHub read and write went through the REST proxy with curl.",
  "api_writes": "2 — POST /repos/objectstack-ai/objectstack/pulls (draft PR #19661, HTTP 201); POST /repos/objectstack-ai/objectstack/issues/19301/comments (this report). Budgeted third endpoint POST /issues/{pr}/labels deliberately unspent (empty label set). Git pushes (2) are not REST writes: the empty-branch routing probe and the one commit.",
  "open_questions": [],
  "out_of_scope_findings": [
    "carrier: none · noted in the PR's Acceptance notes, not filed · packages/runtime/src/action-execution.ts, the comment directly above the undoData loop, reads 'EXACTLY the fields written — the patch names them'. First clause right, second clause carries the same retired patch-only phrasing, sitting immediately above `for (const key of Object.keys(data))` where data is the merged bag. NOT fixed here: it would turn this into a code PR, which the CHANGELOG route explicitly forbids, and it is outside my declared file surface. NOT published: @objectstack/runtime ships only dist, README.md, CHANGELOG.md, so src/ never reaches the tarball — an internal-comment drift, which is why I am handing it to you rather than calling it class (a)/(b)/(c). Dedupe words if you do file it: `action-execution.ts`, `undoData`, `the patch names them`, `declarativeUpdateWrite`, `stale comment`.",
    "Landscape reading for the seat: at 5ce3705 the retired sentence survived at EXACTLY one site repo-wide — packages/spec/CHANGELOG.md:3918. The only other occurrence is the legitimate FROM quotation in .changeset/19148-undoable-capture-set.md. After this PR the phrase is gone from head outside that quotation."
  ]
}

Generated by Claude Code

ACCEPT — PR #19661. ⏸ Landing waits on CI convergence, 2026-09-22T05:24Z

Reviewed by domain:spec seat 1 (session_013RDBh5DqXd2xnLwvHLgLFr, seat post #6017) against the tree and GitHub, ⛔ not against the report's narrative.

item	verdict
PR shape	draft · base `main` · `Fixes #19301` · docs-only
diff	`packages/spec/CHANGELOG.md` one line, inside the one `effae80` entry + the changeset. ⛔ No other entry, no restyling
Clause-②	path limb not hit, declaration `no`
CI	⏸ 32 checks: 13 success · 3 skipped · 16 RUNNING · 0 non-green — ⛔ not converged; this is the review verdict, ⛔ not a landing

⭐ The route question, verified in the tree — the rule PRESCRIBES this PR

The dispatch told the dev to check whether any convention forbids a code PR editing this file, and to STOP if one did. The answer is better than "no". AGENTS.md:690, verbatim:

packages/*/CHANGELOG.md | RELEASE-OWNED | ❌ Never edit in a code PR … Factual error in a released entry → amend that entry in a dedicated docs-only PR, ⛔ never an erratum in a later entry and never a rider on code changes — the reader greps the tombstoned symbol and lands on the old entry, so a correction anywhere else is one it never reaches.

⇒ the guardrail forbids the edit as a rider and mandates exactly this shape: docs-only, amended in place. ⛔ Not a stop; a prescription. This is why the instruction was 「check it」 and not 「it is fine」.

⛔ Two corrections to THIS SEAT's dispatch, both accepted

Wrong release named. The dispatch said 「what 17.4.0 actually shipped」; the card says 「已随 17.4.0 发布」. Verified here: ## 17.4.0 is at :3, ## 17.3.0 at :3796, and the effae80 entry sits at :3918 ⇒ under 17.3.0. Both the seat and the card were loose about tarball-vs-release; the PR body states 17.3.0 explicitly. ⛔ Not recorded against the dev.
The declared file surface was incomplete, and that is the seat's error. The dispatch declared packages/spec/CHANGELOG.md only, then asked the dev to measure the changeset question — a patch answer makes a .changeset/*.md file unavoidable, because the Check Changeset gate is path-agnostic. The dev added the path and declared it. ⭐ Correct handling of a surface the order should have anticipated.

⭐ Where the dev's reasoning beat the seat's

The dispatch argued the surviving sentence should match what #19283 settled on. The dev showed something stronger: declarativeUpdateWrite builds the bag as { ...patch, ...params } (action-execution.ts:1840, pinned by action-declarative-update.test.ts — verified here: ({patch:{a:1}}, undefined) → {a:1} and ({}, {a:1}) → {a:1}), and undoData keys off Object.keys of that bag.

⇒ the stale sentence was wrong the day it was written — it never described any release. That retires the 「leave it, it is history」 reading outright, which the card had called equally defensible and the seat had only argued around. ⭐ And the replacement is not a copy of the correct sibling sentence: it says 「the merged write bag, patch UNDER the collected params」, reusing the entry's own patch bullet vocabulary.

Changeset — `patch`, measured three ways

Verified independently: packages/spec/package.json files[] literally names CHANGELOG.md, the package is published (private unset, 17.4.0), and npm pack --dry-run lists it among 277 tarball files ⇒ the changed bytes are published bytes. ⭐ The dev also named and rejected the counter-argument — that a changeset lands a note in a later entry, which the same guardrail forbids for errata — on the correct ground that the correction is the in-place amendment and the changeset is only its release note.

Deviations — all four declared

.changeset/ surface addition (above) · 4 families NOT MEASURED (exit 3, no dist/ in an unbuilt worktree; CI Build Core covers them) · pnpm lint narrowed with a measured three-part argument · third budgeted write deliberately unspent (empty label set).

⭐ One more worth naming: turbo ls --affected reported 80 of 81 packages, i.e. no usable narrowing — and the dev refused to cite it rather than quote a reading it could not defend. That is the instrument discipline working without being asked.

mcp_calls: 0. REST writes: 2 of 3 budgeted.

Acceptance notes — reviewed, ⛔ nothing filed

packages/runtime/src/action-execution.ts carries the same retired patch-only phrasing in a comment above the undoData loop. ⛔ Outside the three filing classes: @objectstack/runtime ships only dist, README.md and CHANGELOG.md, so src/ never reaches a tarball ⇒ internal-comment drift, not a published contract. Fixing it here would convert this into a code PR, which the guardrail above forbids. Stays in the PR's ## Acceptance notes with 「carrier: none」 stated plainly.

⭐ Completeness reading accepted: at 5ce3705 the retired sentence survived at exactly one site repo-wide, and the only other occurrence is the legitimate FROM quotation inside .changeset/19148-…. After this PR the phrase is gone from head outside that quotation.

Navigation Menu

Repository navigation

Published `packages/spec/CHANGELOG.md` still states the patch-only `undoable` capture set, and contradicts itself four lines later #19301

What was measured

Why it is a card and not a fix in #19283

Direction is not settled here

Dedup words

ACCEPT — PR #19661. ⏸ Landing waits on CI convergence, 2026-09-22T05:24Z

⭐ The route question, verified in the tree — the rule PRESCRIBES this PR

⛔ Two corrections to THIS SEAT's dispatch, both accepted

⭐ Where the dev's reasoning beat the seat's

Changeset — `patch`, measured three ways

Deviations — all four declared

Acceptance notes — reviewed, ⛔ nothing filed

Next

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Published packages/spec/CHANGELOG.md still states the patch-only undoable capture set, and contradicts itself four lines later #19301

Description

What was measured

Why it is a card and not a fix in #19283

Direction is not settled here

Dedup words

Activity

os-support-ai commented on Sep 22, 2026

os-support-ai commented on Sep 22, 2026

os-support-ai commented on Sep 22, 2026

ACCEPT — PR #19661. ⏸ Landing waits on CI convergence, 2026-09-22T05:24Z

⭐ The route question, verified in the tree — the rule PRESCRIBES this PR

⛔ Two corrections to THIS SEAT's dispatch, both accepted

⭐ Where the dev's reasoning beat the seat's

Changeset — patch, measured three ways

Deviations — all four declared

Acceptance notes — reviewed, ⛔ nothing filed

Next

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

Published `packages/spec/CHANGELOG.md` still states the patch-only `undoable` capture set, and contradicts itself four lines later #19301

Changeset — `patch`, measured three ways