qa(checklist): re-point eight stale checklist items across five cards to the product at e540baf8 by objectstack-fleet[bot] · Pull Request #33273 · objectstack-ai/objectstack

objectstack-fleet · 2026-11-14T15:16:30Z

Fixes #33032
Fixes #33054
Fixes #33060
Fixes #33086
Fixes #33091

Clause-②: no
Metadata-protocol: none

维护者速读

改了什么:平台测试清单里 8 个条目的验收文字改成与产品现状一致(唯一键 409 现在报出字段、元数据开关路由数 4/14、内部字段测试文件新路径、os validate 的 controlled_by_parent 分支与新的 os lint 子句、Studio「未运行」原因的显示位置、非参与者伪造审批得 404、仍按部署唯一持有的类型从 webhook 改为 connector),每条都升了 revision 并补了 history 行。
为什么改:这些条目是产品对、清单文字落后,QA 每轮都会把它们误判为失败;不改产品行为。
风险与回滚:只改 docs/qa/platform-checklist/areas/ 下 6 个 JSON 文件,不发布、不进运行时;回滚即 revert 本 PR。
先看:areas/platform-core.json(bare-name 条目改指 connector,并改为按 NAMESPACE_SCOPED_BARE_NAME_TYPES 的补集取类型)、areas/cli.json(新增 acceptance[3] 的 os lint 子句)、areas/approvals.json(acceptance[4] 的 404/403 分界)。
两分钟验证:pnpm check:platform-checklist,预期最后一行 check-platform-checklist: OK — 16 areas, 427 items,symbol anchors 为 1635/1645 resolved。
需要你做的:无,席位落地。

What changes

One commit per card; every edited item bumps revision and appends a history row naming its card and the base it was read at (e540baf8).

Card	Item	Rev	What moved
#33032	`records-forms.field-unique-enforcement`	13 to 14	acceptance[0] and automated.ref: the bare `unique: true` collision names the field ('A record with this SKU already exists', `field` 'sku'), as `28a1527` made it
#33032	`api-backend.error-sentences-and-unique-violation-wire-code`	8 to 9	acceptance[2] and automated.ref: details exactly `{developerMessage, field, object}`, field = the duplicated column. This is also the trailing note on #33086, done once here
#33032	`ai.mcp-read-file-pdf-word-extraction`	2 (unchanged)	no edit: `29f9d3c` (#33021) already landed r2 (60 ms budget, the zero-page shape as a fixture signal, `showcase_invoice_line.receipt`) after the reading the card came from
#33086	`api-backend.rest-metadata-config-contract`	9 to 10	`maintenance` removes 4 routes (adds `POST {prefix}/_migrate-record-currency-codes`), `item` removes 14 (adds `GET {prefix}/:type/:ns/:name`); steps[5] to [7], acceptance[2], two variant rows. The duty to enumerate each switch's radius from a route-table diff (the #15542 ruling) is kept
#33086	`api-backend.second-object-exposure-and-internal-fields`	3 to 4	automated.ref: the moved position pin `packages/objectql/src/protocol-data-door-internal-field-positions.test.ts` plus the delegation twin in `packages/metadata-protocol`
#33086	`cli.validate-v18-authoring-verdicts`	2 to 3	the cbp arm plants the shapes the parse refuses (no master_detail; two required ones, inline and through objectExtensions); the optional master is a control; the lint-only shape moves to a new `os lint` clause, acceptance[3]
#33054	`studio-authoring.flow-status-matches-runtime`	5 to 6	acceptance[1]: the rail and header title carry the worded `unboundCause` (else the reason), the reason verbatim is the Problems note's detail; negative[] untouched (ruling 6545938831 item ①)
#33060	`approvals.record-page-decisions`	5 to 6	acceptance[4] and steps[8]: a non-participant who cannot read the request gets 404 `REQUEST_NOT_FOUND`; 403 `FORBIDDEN` stays the answer to a reader with no slot
#33091	`platform-core.bare-name-one-holder-refusal`	3 to 4	acceptance[0] and [3] drive `connector` (the only covered type outside `NAMESPACE_SCOPED_BARE_NAME_TYPES` at this base); webhook joins the per-namespace leg with the `requires: ['webhooks', 'realtime']` fixture note; the driven type is named as that list's complement

No product code, no changeset (nothing here is in any package's published files[]; skip-changeset rides the create). coverage.json and FOLLOW-UPS.md were not touched: the gate neither regenerated nor required them.

Tests

No test file changed. The judging gate and the derived local families, run at head 51098eec:

pnpm check:platform-checklist

check-platform-checklist: OK — 16 areas, 427 items (419 active, 3 planned); coverage: 40 kinds mapped, 4 waived; traps: 19 documented, 19 in use; provisioning: 10 area recipes, 18 item references resolved (5 area-qualified), 10/10 recipes referenced; ... symbol anchors: 1635/1645 resolved by symbol-anchors.mjs ... 10 on the named residual, 18 file floors held

At base e540baf8 the same gate read 1621/1631 resolved: the 14 new anchors this diff cites all resolve, and the residual stays 10.

derived families (node scripts/pm/dispatch-gates.mjs --commands, 16)

All 16 exit 0, each exit captured before any pipe: check-adr-links (+ --self-test), check-closing-keyword-parity (+ --self-test), check-comment-mask-corpus, plan-test-core-shards --self-test, @objectstack/lint check:doc-formula-expressions (exit 3 PREREQUISITE NOT MET on the first run, an unbuilt @objectstack/formula / @objectstack/lint; exit 0 after building those two under the verify lock), check:ci-filter-parity, check:cross-package-test-inputs, check:doc-authoring, check:driver-memory-census, check:gitlink-declared, check:nul-bytes, check:platform-checklist, check:refd-timer-probe, check:watch-hint-literal. dispatch-gates --ran: 16 derived, 16 run, 0 NOT-MEASURED, 0 UNRUN.

Acceptance notes

Two mechanism readings relayed with the dispatch were wrong at this base, and the edits follow what was measured. (1) The internal-field position pin moved in b388cfe (fix(objectql)!: the engine refuses evaluating an internal field for every caller but a system context #29003), read from the commit listing of both paths; bca2a51 (fix(plugin-security, lint): a refused controlled_by_parent write names no master to a writer who cannot read the master reference #32128) touches neither file. (2) tool is no longer held once per deployment: c747cf0 (feat(objectql, core, metadata)!: tool names held per namespace, and the metadata service keeps each package's tool in its own (package, name) slot (ADR-0150 child B) #33124) put it in NAMESPACE_SCOPED_BARE_NAME_TYPES after QA's control at 8e3d4a5, so the one covered type outside that list is connector, and Re-point platform-core.bare-name-one-holder-refusal acceptance[0] and [3] from webhook to tool: webhooks are held per namespace since #32577 #33091's item drives it.
Read from source and pins, not driven live by this PR: the connector install refusal and cold-boot warning (held by the connector control cases of packages/objectql/src/registry-bare-name-one-holder-flow-family.test.ts), and the new os lint clause of the cli item (its rule's reach note in packages/lint/src/data-model-rules.ts and its unit pin). The os validate / os build exit 0 on the omitted-required shape is QA's live reading. The next QA run scores these.
Re-point approvals.record-page-decisions acceptance[4]: a forged decision by a caller who cannot read the request answers 404 REQUEST_NOT_FOUND, not 403 FORBIDDEN #33060's sibling readings (not edited here, for the seat to carry). The approvals door answers 404 REQUEST_NOT_FOUND to anyone who cannot read the request (ApprovalService.refuseUnreadableRequest, asked by decide, recall, send back, resubmit, reassign, remind, request info and comment). Participants (submitter, current approver, past actor) and the override actor can read it.
- approvals.request-page-reply-and-reader-visibility acceptance[1] (rev 5): stale. The org member, who is neither approver nor submitter on a showcase request, cannot read it (no showcase object declares enable.approvalsVisibleToReaders), so the comment POST answers 404 REQUEST_NOT_FOUND, not 403 FORBIDDEN.
- approvals.quorum-m-of-n steps[4] (rev 9): half stale. The repeat-actor probe by Ada (a past actor, so a reader) answering 403 holds. "A user holding none of the routed positions is refused the same way" does not: that non-participant gets 404.
- approvals.decision-action-matrix acceptance[6] (rev 10): under-specified. "Recall attempted by a non-submitter is FORBIDDEN" holds for a non-submitter who can read the request (a pending approver), but a non-participant gets 404. The clause names no persona. negative[2] (N2) is not stale: it judges acceptance, not the refusal code.
api-backend.rest-metadata-config-contract: the item switch's describe() (packages/spec/src/api/rest-server.zod.ts) still omits GET /meta/:type/:ns/:name, though the mount table gates it. The maintenance describe() is current. checklist: re-point three api clauses that lag the product at 8e3d4a54 (QA 18.4.0 R2 part B, B-api slice) #33086 names this as its own QA card (card-B-api-5); this PR leaves the spec alone.

Verification

pnpm check:platform-checklist
node scripts/pm/dispatch-gates.mjs --commands    # 16 families, all run green at 51098eec

Session: session_017xRBcRZn9wGkMSrznVX4JS (seat domain:devx#2).

Generated by Claude Code

…l (card 33032) records-forms.field-unique-enforcement r14: acceptance[0] and automated.ref expect the bare `unique: true` collision on the live SQLite leg to name the field ('A record with this SKU already exists', developerMessage '... a unique constraint on 'SKU' already holds this value.', `field` 'sku'), as 28a1527 made it and the dogfood pin already asserts. api-backend.error-sentences-and-unique-violation-wire-code r9: acceptance[2] and automated.ref expect details exactly {developerMessage, field, object} with field = the duplicated column. Item 3 of the card (ai.mcp-read-file-pdf-word-extraction acceptance[3]) is already satisfied at this base: 29f9d3c landed its r2 (60 ms budget, the zero-page shape named as a fixture signal, showcase_invoice_line.receipt) after the reading the card was drafted from. No edit. Claude-Session: https://claude.ai/code/session_017xRBcRZn9wGkMSrznVX4JS Co-authored-by: Claude <noreply@anthropic.com>

…e product (card 33086) api-backend.rest-metadata-config-contract r10: the `maintenance` radius is four (POST {prefix}/_migrate-record-currency-codes joined it) and `item` is fourteen (GET {prefix}/:type/:ns/:name rides the item read), as packages/rest/src/rest-config-mount-table.pin.test.ts measures. steps[5] to [7], acceptance[2] and the two variant rows re-counted; the duty to enumerate each switch's radius from a route-table diff is kept. api-backend.second-object-exposure-and-internal-fields r4: automated.ref names the moved position pin in packages/objectql and the door's delegation twin in packages/metadata-protocol, in place of a file that no longer exists. cli.validate-v18-authoring-verdicts r3: the controlled_by_parent arm plants the shapes the parse refuses (no master_detail; two required ones, inline and through objectExtensions); the optional master is a control; the lint-only shape (an omitted `required` on the only master_detail) moves to a new `os lint` clause, acceptance[3]. The card's fourth note, error-sentences acceptance[2], is the same clause as item 2 of card 33032 and was re-pointed once, in that card's commit. Claude-Session: https://claude.ai/code/session_017xRBcRZn9wGkMSrznVX4JS Co-authored-by: Claude <noreply@anthropic.com>

…unning cause (card 33054) studio-authoring.flow-status-matches-runtime r6: for a known unboundCause the rail and header title read the worded cause (on a default boot, deployment_policy), and the platform's reason verbatim is the Problems note's detail, which describeFlowRunStatus does on purpose. acceptance[1], steps[3], the title, automated.ref (adds the runStatus unit pin) and source[0] follow; the note still counts as no problem, and negative[] is untouched. Claude-Session: https://claude.ai/code/session_017xRBcRZn9wGkMSrznVX4JS Co-authored-by: Claude <noreply@anthropic.com>

…4 a non-reader's forged decision gets (card 33060) approvals.record-page-decisions r6: the step's contributor-bound member can read the project but not its request, and every door that acts on a request answers such a caller REQUEST_NOT_FOUND, exactly as a missing id (ApprovalService.refuseUnreadableRequest), so the forged approve answers 404 and the request stays pending. 403 FORBIDDEN stays the answer to a reader who holds no slot. steps[8], acceptance[4] (clause, verify, evidence) and source[5] follow; negative[2] is unchanged. Claude-Session: https://claude.ai/code/session_017xRBcRZn9wGkMSrznVX4JS Co-authored-by: Claude <noreply@anthropic.com>

…connector (card 33091) platform-core.bare-name-one-holder-refusal r4: webhook (e55347d) and tool (c747cf0) both joined NAMESPACE_SCOPED_BARE_NAME_TYPES, so at this base the one covered type still held once per deployment is connector. acceptance[0] and [3], the title, fixtures, steps[1], negative[0], the variant and source[4] now drive a connector catalog descriptor qa_conn; webhook joins the per-namespace leg of acceptance[1] beside flow and agent, with the fixture note that a webhook-declaring package declares requires: ['webhooks', 'realtime']. acceptance[0] names the driven type as the complement of NAMESPACE_SCOPED_BARE_NAME_TYPES within the covered types, so the next narrowing reads as a re-check rather than a stale FAIL. Claude-Session: https://claude.ai/code/session_017xRBcRZn9wGkMSrznVX4JS Co-authored-by: Claude <noreply@anthropic.com>

objectstack-fleet · 2026-11-14T15:22:03Z

domain:devx#2 · sunhaolin · session_017xRBcRZn9wGkMSrznVX4JS

席位核对 · ACCEPT · 2026-11-14T15:21Z · head `51098eec`

席位意见：文字对齐产品，不改产品行为；可由席位落地（非受管面、+101/−51，未达人工合并阈值）。

Verdict: ACCEPT, checked against GitHub and origin/main e540baf8, not against the report (6879560259).

Shape. Draft PR against main. Lines 1–5 of the body read Fixes #33032, Fixes #33054, Fixes #33060, Fixes #33086 and Fixes #33091, one per fold member, each merge meant to close its card. The other closing-keyword hits in the body point at closed or merged numbers only. There are five commits, one per card (549d5a6, ef78485, 5505803, c6c56d4, 51098ee). Clause-②: no and Metadata-protocol: none stand on their own lines. With none and no narrowing arm, no Skills-docs: line is owed. The path face from pulls/33273/files is six files, all under docs/qa/platform-checklist/areas/ (api-backend +31/−13, approvals +8/−7, cli +20/−7, platform-core +18/−12, records-forms +10/−4, studio-authoring +14/−8), inside the claimed surface. No governed path. skip-changeset fits a docs-only QA diff.
Premises, measured by the seat:
- packages/core/src/package-scoped-resolution.ts at e540baf8: BARE_NAME_ONE_HOLDER_TYPES minus NAMESPACE_SCOPED_BARE_NAME_TYPES is connector alone, since tool joined the second list in c747cf0. So the bare-name item's move to connector, naming the list's complement, is right.
- refuseUnreadableRequest is present in approval-service.ts (7 hits).
- ai.mcp-read-file-pdf-word-extraction is already rev 2 with the 60 ms budget at base, so leaving it unedited is right.
- Read in the diff: the approvals 404/403 split, and the platform-core history row that cites feat(plugin-webhooks)!: declared webhooks are held per namespace, each holder dispatched and switched as its own (ADR-0150 child B, webhook family 2/3) #32577 and feat(objectql, core, metadata)!: tool names held per namespace, and the metadata service keeps each package's tool in its own (package, name) slot (ADR-0150 child B) #33124.
Gates. Check-runs on 51098eec at this comment: 14 success, 23 skipped, 0 failed, 8 in progress (Test Core 1–3/8, the five Lint jobs, both Type Check jobs). Landing waits until every check is green or an expected skip. Locally the dev ran 16 of 16 derived families, check:platform-checklist exit 0, dispatch-gates --ran 0 UNRUN. Those runs are local and are not counted for CI.
Accepted deviations.
- The seat's own corrections, owned here. The dispatch's Zone 2 claimed tool was still one holder per deployment, and that bca2a51 moved the internal-field pin. Both were wrong at base. The claim's Premise: line also listed read-file rev 2 as stale, and it was already fixed by qa(checklist): correct five assertion defects from the 18.4.0 acceptance #33021. The dev measured and re-pointed correctly.
- The pin-move commit b388cfe (fix(objectql)!: the engine refuses evaluating an internal field for every caller but a system context #29003) comes from the dev's REST commit listing. The seat's clone is shallow, so the seat did not re-measure it.
- The connector and os lint legs were read from source and unit pins, not driven live; the Acceptance notes say so. The next QA run on these items is the live check.
- The commit trailers follow AGENTS.md.
Out of scope.
- The three approvals siblings (request-page-reply A1, quorum S4, decision-action-matrix A6) were not edited here → filed checklist: re-point three approvals clauses that still expect 403 where a non-reader gets 404 REQUEST_NOT_FOUND (sibling half of #33060) #33275, a sub-issue of Re-point approvals.record-page-decisions acceptance[4]: a forged decision by a caller who cannot read the request answers 404 REQUEST_NOT_FOUND, not 403 FORBIDDEN #33060, p2.
- The item describe() lag in rest-server.zod.ts → carrier: qa#3's card-B-api-5, as named in checklist: re-point three api clauses that lag the product at 8e3d4a54 (QA 18.4.0 R2 part B, B-api slice) #33086's body (domain:spec). The seat files nothing here.
Landing. Once every check is green, one relay stroke: pr_ready + automerge_enable. Merging closes the five cards; the seat then clears pm:dispatched and follows the PR to MERGED.

objectstack-fleet · 2026-11-14T15:30:20Z

domain:devx#2 · sunhaolin · session_017xRBcRZn9wGkMSrznVX4JS

Landing · 2026-11-14T15:30Z · head `51098eec`

Gates at this head: every check-run is green or an expected skip: 28 success, 23 skipped, 0 failed, 0 running. Lint & Repo Gates and TypeScript Type Check both read success.
K: 1 first-parent commit on main since the PR base d4fbb4db, under the 15 that would require merging main first. No generated files are in the diff.
Verdict on record: ACCEPT 6879625279, same head. Non-governed, +101/−51.
Stroke: pr_ready + automerge_enable. The merge closes checklist: re-point two api clauses to the shape #32935 landed, and fix the read-file CPU-budget step (QA 18.4.0 R2 api slice) #33032, Re-point studio-authoring.flow-status-matches-runtime acceptance[1]: a known "not running" cause is worded in the rail and header titles, and the platform reason is the Problems note's detail #33054, Re-point approvals.record-page-decisions acceptance[4]: a forged decision by a caller who cannot read the request answers 404 REQUEST_NOT_FOUND, not 403 FORBIDDEN #33060, checklist: re-point three api clauses that lag the product at 8e3d4a54 (QA 18.4.0 R2 part B, B-api slice) #33086 and Re-point platform-core.bare-name-one-holder-refusal acceptance[0] and [3] from webhook to tool: webhooks are held per namespace since #32577 #33091. The seat follows this PR to MERGED and then clears pm:dispatched on all five.

claude added 5 commits November 14, 2026 15:00

objectstack-fleet Bot assigned sunhaolin Nov 14, 2026

objectstack-fleet Bot added the skip-changeset PR has no user-facing published change; bypasses the changeset gate label Nov 14, 2026

github-actions Bot added the size/m label Nov 14, 2026

objectstack-fleet Bot mentioned this pull request Nov 14, 2026

checklist: re-point three approvals clauses that still expect 403 where a non-reader gets 404 REQUEST_NOT_FOUND (sibling half of #33060) #33275

Closed

objectstack-fleet Bot marked this pull request as ready for review November 14, 2026 15:30

objectstack-fleet Bot enabled auto-merge November 14, 2026 15:30

objectstack-fleet Bot added this pull request to the merge queue Nov 14, 2026

This was referenced Nov 14, 2026

[PM seat] domain:devx · seat 4 — 🟢 os-project-manager · p1 + p2 #33288

Open

metadata.endpoints.item's describe() omits the qualified read GET /meta/:type/:ns/:name that the switch also removes (low) #33085

Closed

Merged via the queue into main with commit fdb76fb Nov 14, 2026
52 checks passed

objectstack-fleet Bot deleted the claude/issue-33032-checklist-repoint-fold branch November 14, 2026 15:57

Navigation Menu

Repository navigation

qa(checklist): re-point eight stale checklist items across five cards to the product at e540baf8 - #33273

qa(checklist): re-point eight stale checklist items across five cards to the product at e540baf8#33273
objectstack-fleet[bot] merged 5 commits into
mainfrom
claude/issue-33032-checklist-repoint-fold

objectstack-fleet Bot commented Nov 14, 2026

Uh oh!

objectstack-fleet Bot commented Nov 14, 2026

Uh oh!

objectstack-fleet Bot commented Nov 14, 2026

Uh oh!

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

objectstack-fleet Bot commented Nov 14, 2026

维护者速读

What changes

Tests

Acceptance notes

Verification

Uh oh!

objectstack-fleet Bot commented Nov 14, 2026

席位核对 · ACCEPT · 2026-11-14T15:21Z · head 51098eec

Uh oh!

objectstack-fleet Bot commented Nov 14, 2026

Landing · 2026-11-14T15:30Z · head 51098eec

Uh oh!

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

席位核对 · ACCEPT · 2026-11-14T15:21Z · head `51098eec`

Landing · 2026-11-14T15:30Z · head `51098eec`