[finding] `resolve-authz-context.ts` §6b-config states a rule a maintainer ruling REVERSED, and cites a pin that now records the reversal — the first thing the next implementer of this site reads

[os-project-manager]

A load-bearing comment in packages/core/src/security/resolve-authz-context.ts asserts a behaviour that a maintainer ruling reversed, and cites as its authority a pin that now says the opposite of what the comment claims it says.

Surfaced by the domain:services seat (#6021) while adjudicating #18336. ⚠️ The measurement below is this seat's own — the delivering dev reported this finding with a mechanism that does not hold, and the corrected reading is recorded here instead. Grading and routing left to triage.

The comment

origin/main:packages/core/src/security/resolve-authz-context.ts :1005–1007, verbatim:

// nothing, because the single promotion is pinned NEVER to consult that
// variable (bootstrap-platform-admin-walled-owner.test.ts, "never
// consults the owner-email variable"). That holds whichever way Choice 4B

Why it is false

Maintainer ruling 2026-09-08, decision batch #100, recorded on #16682 (comment 5587754690, author os-zhuang), verbatim:

F3 — the Choice 4A sentence is superseded for this one point. Under single posture the first-boot promotion consults OS_PLATFORM_OWNER_EMAIL first.

⇒ the comment states the pre-reversal rule as current fact.

⚠️ Correcting the mechanism — the cited pin DOES still resolve

The dev report on #18336 stated that 「the test of that name no longer exists」. That is wrong, measured here with controls:

probe	reading
never consults the owner-email variable in bootstrap-platform-admin-walled-owner.test.ts	1
positive control (the \group` posture is walled too`) in the same file	1
same needle, repo-wide over packages/	1 — that file only
control for the same corpus (owner-email variable)	4 files ⇒ the instrument is live, the 1 is real

The string survives as an explicitly-superseded quotation, inside a block at :434–456 that opens ⚠️ RE-AUTHORED by #16682 and states in terms that the case 「used to assert the opposite」, naming the maintainer ruling as the authority for the reversal.

⇒ The test file is correct and well-documented. The defect is one-sided: resolve-authz-context.ts cites that pin as if it were live support for a claim the pin itself now records as overturned. A reader who follows the citation lands on text that contradicts the comment that sent them there.

Why it is worth a card

⭐ It is the first thing the next implementer reads. The comment is the stated justification for the posture gate on the deprecation notice, and it sits at the exact derivation site that #18336 (and, downstream, #11979 / Choice 4B) has to modify. #18336 is now in the decision box partly because the posture semantics around this site are contested — a comment asserting a reversed rule at that site is an active hazard, ⛔ not cosmetics.

⚠️ Scope note, ⛔ not a proposed remedy: the neighbouring sentence 「That holds whichever way Choice 4B [is decided]」 may also be affected, since it is conditioned on the same superseded premise. Whoever takes this should judge the whole comment block, ⛔ not just the one clause.

Dedup words: stale code comment · single promotion consults owner email · #16682 F3 supersession · resolve-authz-context 6b-config comment · platform admin re-anchor comment drift

Measured at 2026-09-16T06:54Z against origin/main.

---

Generated by Claude Code

[os-sam]

Triage 定级(常规档) · 2026-09-17T10:42Z · 分诊席 objectstack#6015 · R+269b

• 判据 (b) —— packages/core/src/security/resolve-authz-context.ts:1005–1007 的承重注释断言一条维护者裁决已经推翻的行为,并引用一条如今记录着该推翻的 pin 作为权威。裁决原文(2026-09-08,决策批 Add copy entire page as markdown feature to documentation #100,plugin-security: the first-user promotion picks the oldest authenticable user from an UNORDERED 50-row sys_user window, so on the default driver a seeded job seeker became platform admin and owned every seeded row #16682 评论 5587754690):「F3 —— Choice 4A 那句在这一点上被取代;single 姿态下首次启动提升先咨询 OS_PLATFORM_OWNER_EMAIL」。⇒ 注释把推翻前的规则当现行事实陈述。
• 车道 domain:engine(packages/core)· priority:p2 · type Bug。
• ⭐ 为什么不是 p3「只是注释」:它是下一个改这个站点的人第一眼读到的东西,而这个站点是权限提升。一条把已推翻规则说成现行的承重注释,是在给下一次改动预装一个错误前提。
• ✅ 填报席申报:交回这条的 dev 给的机制不成立,这里记的是该席自己重量的读数 —— ⛔ 不是转述。

⛔ 不认领、不派发。
_{本评论来自分诊座位}

---

Generated by Claude Code

[huangyiirene]

Claim: PM loop round 1
Session: session_01CqmCgU5RGDoJYhHUMVp2af
Branch: claude/issue-18380-authz-context-6b-config-comment
Worktree: objectstack-issue-18380
Domain: domain:engine
Seat: domain:engine#1
File surface: packages/core/src/security/resolve-authz-context.ts (stop on breach; explain in the report)
Container & model: S, not mechanical, mode:subagent, model: opus (default judgement tier) — quoting this dispatch's own dispatch-gates.mjs --tier: 「Model tier — no path-derived mandate … The tier stays the PM's per-card judgment call (floor sonnet · default opus · ceiling fable).」 ⇒ ⛔ not the floor: a load-bearing comment that cites an authority is being re-keyed against a maintainer ruling, and getting the citation wrong a second time is the failure mode this card exists to repair.
Clause-②: no
Thread-read: 5713023506
Serial constraints cleared: In-flight devs in this lane: ZERO at claim time. Three sibling PRs are in the MERGE QUEUE and all are disjoint from this file — #18630 (packages/drivers/driver-sql/), #18634 (packages/metadata-protocol/), #18637 (packages/objectql/), each read from get_files, ⛔ not from a report. Nothing in flight or queued touches packages/core/src/security/. Verify lock at claim time: free, queue empty ⇒ arrival depth 1, under LOCK_DEPTH_HOLD.

Premises re-measured by this seat on origin/main at claim time

premise	reading
the false comment is still there	✅ resolve-authz-context.ts:1005-1007, verbatim: 「the single promotion is pinned NEVER to consult that variable」
the cited pin still resolves	✅ 1 hit repo-wide for 「never consults the owner-email variable」
⭐ firing control	「owner-email variable」 hits 4 files ⇒ the 1 above is a reading, ⛔ not a dead grep

⇒ The card is live, and its correction of the original dev's mechanism is right: the pin did not disappear. ⭐ Worth keeping, because it is the whole reason this card reads the way it does — the first report said 「the test of that name no longer exists」, and that was measured false by the filing seat with controls. ⛔ Do not re-derive it from the first report's version.

⚠️ A path correction this seat owes you, made while verifying

The card and the comment both cite the pin by bare filename, which is how the comment spells it too. This seat first looked for it under packages/core/src/security/ — it is not there. Measured location:

packages/plugins/plugin-security/src/bootstrap-platform-admin-walled-owner.test.ts

⇒ ⛔ The card is not wrong; this seat's directory assumption was, and it is recorded so you do not lose ten minutes to the same guess.

⛔ That file is in ANOTHER LANE. Read it; do not edit it.

packages/plugins/plugin-security/** is domain:services by the lane table. It is outside your file surface and outside this lane.

• ✅ Read it — you must, to state what the pin actually records now.
• ⛔ Do not edit it, even if its own superseded-quotation block also reads oddly. If you conclude it needs a change, report that; a follow-up card in the owning lane is this seat's to file, ⛔ not yours to ride in on this PR.

⭐ Note the shape you are repairing, because it is subtle: the pin's string survives as an explicitly-superseded quotation inside a block marked 「RE-AUTHORED by #16682」. So the comment's citation resolves to text that now says the opposite of what the comment claims it says. ⇒ 「the citation still resolves」 and 「the citation supports the claim」 are two different questions, and only the first is true.

---

Generated by Claude Code

[huangyiirene]

os-dev-report

{
  "issue": 18380,
  "status": "done",
  "branch": "claude/issue-18380-authz-context-6b-config-comment",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/18645",
  "session": "session_01CqmCgU5RGDoJYhHUMVp2af",
  "premise_still_valid": true,
  "files_changed": ["packages/core/src/security/resolve-authz-context.ts"],
  "summary": "Comment-only repair of the §6b-config posture-gate justification. R1 and R2 both re-measured and confirmed on origin/main 32be735e5b: the false comment was present (located by content, not line number), the needle 'never consults the owner-email variable' returns 1 hit repo-wide in packages/plugins/plugin-security/src/bootstrap-platform-admin-walled-owner.test.ts:436, firing control 'owner-email variable' hits 4 files. R3 read at the source: the pin's string survives only inside the ':434-456' block marked 'RE-AUTHORED by #16682', which quotes the maintainer ruling and is followed by the case 'a declared owner DOES redirect the single-org promotion (#16682)'. The comment is re-keyed to post-ruling behaviour, the citation re-pointed at what the pin records now, and the gate's justification restated from what the reversal did NOT change (the declared-owner leg mints the same unscoped admin_full_access row via the single promote() call site at bootstrap-platform-admin.ts:867, and a rig already holding one returns already_have_admin at :687 before that leg runs). The neighbouring Choice 4B clause was checked and was ALSO stale: '4B is ruled as the sequenced follow-up, not dropped' per #11663 comment 5404675670, so 'whichever way 4B is eventually ruled' is replaced by 'ruled and filed, not landed'. No logic, no assertion, no behaviour touched; the plugin-security test file was read only.",
  "line_budget": "n/a — no skills/** path in the diff, no ratchet governs a source comment. For scale: the edited comment block went from 15 to 36 lines (32 insertions / 11 deletions, one file).",
  "gates": "ONE derivation + ONE reconciliation, on final head cb602a03b3: 'node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack --ran RANFILE' exit 0 — 47 derived, 45 run, 2 NOT-MEASURED, 0 UNRUN. 44 of 45 green. RED: 'pnpm check:cross-package-test-inputs' exit 1 = the known gate defect #18621, ⛔ not this PR's red — control measured here: with packages/spec/dist moved aside the same command exits 0 ('OK: 29 package(s) read outside themselves, all declared'), restored afterwards with 216 files before and 216 after. NOT MEASURED: 'pnpm check:dual-build-cjs-loads' and 'pnpm check:lean-entry-closure', both exit 3 PREREQUISITE NOT MET (they load a whole-repo dist), declared to CI. Exit codes were redirected to disk and captured before any pipe.",
  "tests": "pnpm --filter '@objectstack/core...' build -> os-verify-lock VERDICT command-exit 0 (held 131s, waited 4s). pnpm --filter @objectstack/core typecheck -> exit 0 (tsc --noEmit + tsconfig.examples + check:test-typecheck OK). pnpm --filter @objectstack/core test -> exit 0, 'Test Files 51 passed (51) / Tests 1316 passed (1316)'. Repo-wide 'pnpm lint' (eslint . --no-inline-config) run in FULL -> exit 0, so no narrowing had to be declared. Both heavy runs went through scripts/pm/os-verify-lock.sh with OS_VERIFY_LOCK_SLOT=os-dev-18380. No ablation: the diff is comment-only, so there is no behaviour to mutate.",
  "changeset": "none, skip-changeset label applied and read back. Measured, not assumed: @objectstack/core ships files ['dist','README.md','CHANGELOG.md']; after building, the needle text from this comment scores 0 hits across those paths, a second needle taken from a pre-existing body comment in the same function also scores 0, and the positive control (a docblock on the exported hasPlatformAdminStanding) scores 1 in dist/index.d.ts and 1 in dist/index.d.cts. Control lands, needles do not ⇒ a function-BODY comment reaches no published artifact (declaration emitter carries declaration-level docblocks only, esbuild drops body comments, dropSourcesContent keeps source text out of the .map).",
  "mcp_calls": "0 — no MCP GitHub tool was called, read or write.",
  "api_writes": "3 REST proxy writes: POST /repos/objectstack-ai/objectstack/pulls (draft, 201); POST /repos/objectstack-ai/objectstack/issues/18645/labels (additive, 200, read back as ['size/s','skip-changeset'], nothing stripped); POST /repos/objectstack-ai/objectstack/issues/18380/comments (this report). Plus one git push of the branch (empty-branch probe, then the commit). No PATCH of any body.",
  "deviations": [
    "Report field set: the dispatch names nine machine-checkable fields, the role file mandates issue/status/branch/pr/session/summary/tests/mcp_calls/api_writes. Reported as the UNION rather than silently picking one — flagged here rather than choosing a side.",
    "Commit trailer uses the model-free 'Co-authored-by: Claude' spelling per AGENTS.md and check:commit-card-trailers, not the harness attribution reminder's model-bearing form. The pre-push gate confirmed: '1 commit message(s) on this push carry no card relation and no model identifier in the trailer pair.'",
    "check:dual-build-cjs-loads and check:lean-entry-closure were NOT measured (exit 3, whole-repo dist prerequisite). Declared narrowing, left to CI.",
    "needs:contract-review is absent on PR #18645 and was neither added nor removed — seat territory. Reading: 'node scripts/pm/check-clause2-carriers.mjs --pair 18645' exit 0, 'the clause-② declaration is readable in the fixed spelling and both carriers agree, and its diff carries no widening tell'."
  ],
  "open_questions": [],
  "out_of_scope_findings": [
    "noted, not filed: the enclosing describe title in packages/plugins/plugin-security/src/bootstrap-platform-admin-walled-owner.test.ts still reads 'single posture — \"first user is owner\" is ruled reasonable and UNCHANGED (Choice 4A)' while the case inside it now asserts a declared owner redirects the promotion. The block's own re-authored docblock states the split correctly, so this is a title-level nit, not an error a reader can copy into failure — outside the three filing classes. Who will touch it: #11979's implementer, since Choice 4B rewrites exactly this block when it lands. Lane is domain:services; read-only for this card, per the claim."
  ],
  "card_body_integrity": "No sanitizer truncation found in #18380's body: it ends with its intact Claude Code footer, carries no tag-shaped fragment, and both comments (5713023506 triage, 5713709046 PM claim) read complete. The newest Claim: on the card names this branch."
}

---

Generated by Claude Code

[huangyiirene]

os-dev-report reviewed — ACCEPT. PR #18645 goes ready and into the merge queue.

domain:engine execution seat, session_01CqmCgU5RGDoJYhHUMVp2af, R1, 2026-09-17T12:34Z. Report: 5714025771. ⛔ Verified against GitHub and the tree.

Form, scope and CI

criterion	reading
draft · base · head	true · main · cb602a03b3
first body line · Clause-②: no at column 0	both present
changed files	1 — resolve-authz-context.ts, +32/−11, comment block only
path face via get_files	no governed-surface hit
labels	size/s, skip-changeset
CI, latest per name	34 checks · 0 non-green · 0 pending
skipped vs roster	6/6 on EXPECTED_SKIPS
carriers	check-clause2-carriers --pair 18645 exit 0, both agree, no widening tell

R3 was acted on, not merely acknowledged

The dispatch's whole point was that 「the citation resolves」 and 「the citation supports the claim」 are different questions. The dev read the pin at source and established what it records now: the string survives only inside the :434-456 block marked 「RE-AUTHORED by #16682」, which quotes the maintainer ruling and is followed by the case 「a declared owner DOES redirect the single-org promotion」.

⇒ The comment is re-keyed to post-ruling behaviour and the citation re-pointed at what the pin actually records, ⛔ not merely stripped of the false sentence. The gate's justification is restated from what the reversal did not change — the declared-owner leg mints the same unscoped row through the single promote() call site, and a rig already holding one returns already_have_admin before that leg runs.

⭐ The neighbouring clause WAS also stale — the half-fix warning paid for itself

The dispatch said: 「check whether that framing is still accurate before you leave it standing … a comment repaired in one clause and left stale in the next is a half-fix」.

It was stale. Per #11663 comment 5404675670, Choice 4B is 「ruled as the sequenced follow-up, not dropped」 — so 「whichever way 4B is eventually ruled」 was itself false, and is now 「ruled and filed, not landed」.

⇒ ⭐ One card repaired two stale claims because the order asked for the sweep rather than the named line. That is the difference between fixing a defect and fixing a comment.

⭐ The changeset question was MEASURED, and the result generalises

@objectstack/core ships ['dist','README.md','CHANGELOG.md']. After building:

needle	hits in published paths
text from this comment	0
a second needle from a pre-existing body comment in the same function	0
⭐ positive control — a docblock on the exported hasPlatformAdminStanding	1 in dist/index.d.ts, 1 in dist/index.d.cts

⇒ A function-BODY comment reaches no published artifact — the declaration emitter carries declaration-level docblocks only, esbuild drops body comments, and dropSourcesContent keeps source text out of the map. Control lands, needles do not.

⭐⭐ This is the third data point in one round on 「does a comment ship」, and together they form a usable rule: a docblock on an exported symbol ships (#18331, measured in the real tarball); a detached docblock sitting above an exported symbol also ships, because the dts bundler emits both blocks (#18331); a function-body comment does not (this card). ⇒ Future cards in this family can decide the changeset question from where the comment sits, and then verify rather than explore.

Deviations — accepted, and one is a third data point of its own

• ⚠️ Report field set: the dispatch named nine machine-checkable fields; the role file mandates a different set. The dev reported the UNION and flagged it rather than silently picking a side. ⭐ Correct — and this is the third dev this round to flag a divergence between this seat's orders and os-dev.md (the earlier two were the write budget). ⇒ One is an oversight, three is a pattern: this seat's orders have been quietly narrowing or restating role-file clauses. Recorded on the seat post (5713135981); orders now defer to the role file explicitly.
• ⭐ Commit trailer: the model-free Claude spelling was used and the pre-push gate confirmed it. The dispatch's up-front warning saved the bounce every earlier dev paid.
• Two NOT MEASURED families (exit 3, whole-repo dist prerequisite), declared to CI. The known check:cross-package-test-inputs red was named with a control measured in place — with packages/spec/dist moved aside the same command exits 0, restored afterwards at 216 files before and after. ⛔ Not chased, ⛔ not twinned.
• Repo-wide pnpm lint run in full, so ⛔ no narrowing had to be declared.

Findings

⏹️ The describe title in the plugin-security pin still reads 「…UNCHANGED (Choice 4A)」 while the case inside asserts the redirect. Correctly noted, not filed: the block's own re-authored docblock states the split correctly, so it is a title-level nit a reader cannot copy into failure, and the file is domain:services — read-only for this card, per the claim, and the dev respected that. ⭐ Its named successor is #11979's implementer, who rewrites that block when Choice 4B lands.

mcp_calls 0 · REST writes 3, label write read back. ⛔ No denied-roster tool was called.

⇒ Ready + auto-merge (SQUASH). This seat follows it to MERGED.

---

Generated by Claude Code

[github-actions]

os-closed-card-sweep — machine-findable marker for this generated comment.

Removed the pm-loop state label(s) this closed card no longer claims: pm:dispatched.

• Closing pull request: docs(core): re-key the §6b-config posture-gate comment to the ruling it cites #18645, merged.
• Closing commit 84ad2e1394, merged into main.
• Left untouched: priority:p2, domain:engine — ownership, priority and outcome are not state claims.
• The label set was read back after the write and matched.

A state label claims work is in flight. This card is closed on a merged delivery, so the claim
is stale; every other label is left exactly as it was found. Nothing here is a judgement about
the card, and no verdict-bearing label is ever touched by this sweep.

posted by half-state-patrol run 35228917608 · trigger schedule

Generated by Claude Code