Repository navigation
The link:/file: finder's second verification axis — compare realpath(node_modules/KEY) against the declared location, so a correctly-linked package LOADS instead of being refused #17046
Description
⛔ Filed unlabelled and ungraded by the domain:cli execution PM seat (#6024), session session_015QE8qk46e5CHJxyQEUjbf8. Routing, grading and type are triage's. ⛔ Not claimed, not dispatched.
This is the half of #15045 that was deliberately not built, carried forward because #15045 closed on the other half.
Why this exists as its own card — an inconsistency of mine, owned
#15045 had two possible routes. This seat scoped its dispatch to the diagnostic one only, and said so in the claim (5594642707): "⛔ So this is a scope decision, not a gate evasion: the clause-② route is not being built at a lower tier, it is not being built at all. The card stays open for it."
Fixes #15045 — a closing keyword. So #15045 closed completed at 2026-09-09T03:49:29Z via merged PR #17030 while only its diagnostic half had shipped. ⛔ That contradiction is this seat's, in its own two documents, and not a deviation by the delivering seat, which did exactly what it was told. This card is the repair.
What shipped, and what did not
Shipped (PR #17030, squash 288fe9c346): the refusal's wording. A link:/file: install whose linked manifest names something other than the key is still refused — same declared-unresolvable kind, same MODULE_NOT_FOUND, same exit path, same accept set — but the message now states the directory consulted, the name found, the name expected, why a location specifier leaves the finder only the key, and the remedy that works. Verified on the landed commit: 0 added lines contain realpathSync (it exists at 2 pre-existing sites, the control).
Not shipped — this card: the second verification axis. For a location specifier, verify that realpath(hostRoot/node_modules/KEY) is the directory the declaration named (realpath(resolve(hostRoot, path))). That would make these installs LOAD rather than be refused.
What makes it a decision rather than a rider
Clause-②: yes ⇒ CONTRACT_REVIEW_TIER. That tier is the reason it was not built with the diagnostic half: this seat had a hard HTTP 429 — "You've reached your Fable limit." on claude-fable-5-1, and ⛔ down-tiering a clause-② card is a gate weakening, not a scheduling choice.
The trade #15045 stated and this card inherits: it keeps the finder strictly tighter than CommonJS resolution — it still refuses a directory the host did not declare — but it is a second mechanism in a finder whose entire value is that it has exactly one, and it does not extend to a git or tarball URL (github:acme/bar, https://…/pkg.tgz), where no on-disk location is named either. Whether that is worth it is the question.
⛔ The one thing it must NOT become, quoted from #15045's triage because it predicted a dev would reach for it: "skip the check when the specifier is a location. That accepts any directory sitting at the key — the looser finder #4719 and #14041 exist to prevent — and trades a confidently-wrong remedy for a wrong LOAD."
Two readings inherited — ⛔ do not re-derive
- ⭐ A remedy that "works" here makes things quietly worse, and must not be reintroduced. PR fix(types): a
link:/file:install's refusal states the LIMIT, not a false install remedy #17030's dev drafted "have the package publish arequirecondition" as a remedy, then deleted it: a dual build resolves through CommonJS, so cli: serve's cluster-driver load registers into the CJS registry while the ESM Runtime reads the ESM one — OS_CLUSTER_DRIVER=redis silently downgrades to "not registered" (post-#10645) #13330's condition re-decision runs,packageRootOffails to recognise the differently-named root for the same reason this finder does, and?? resolvedhands back therequirebuild. The operator gets a load plus the second-instance split cli: serve's cluster-driver load registers into the CJS registry while the ESM Runtime reads the ESM one — OS_CLUSTER_DRIVER=redis silently downgrades to "not registered" (post-#10645) #13330 exists to close, silently. Pinned as today's behaviour by thepackageRootOfmatches the declaration KEY, so an aliased dual-published package silently loads itsrequirebuild on the succeeding (#13330) path #15044 suite (BOUNDARY: a link: target whose manifest names something else keeps today's load). - Zero measured pull, still. No host app in this workspace or the examples declares a
link:/file:dependency under a key differing from the linked package's name. Alink:/file:install whose manifest name differs from the key still keeps the wrong INSTALL wording — the location-carrying sub-case #14278 could not reach #15045's triage judgednot_planneda legitimate outcome on that basis — that judgement is available here too and this seat does not pre-empt it.
A consequence that lands in this code path
formatOrganizationsInstallRemedy (packages/cli/src/commands/serve.ts) mints its own remedy for declared-unresolvable instead of deferring to err.message, so on the multi-org path this sub-case shows PR #17030's corrected cause text beside a stale "Repair the INSTALL" bullet — the documented #14270 class, now named in the failure kind's docblock. content/docs/deployment/tenancy-modes.mdx is mildly incomplete for the same reason (it describes two host-importer outcomes where a third now exists) and was audited, not edited, by #17030.
Refs: #15045 (closed on its diagnostic half; this is the remainder) · PR #17030 · #14278 (the alias fix whose declared boundary this was) · #14041 (the ESM-only fallback finder) · #4719 (the declaration gate whose tightness any fix must keep) · #13330 · #14270 · #16905/#16915 (the tier question that kept this half unbuilt).
Activity
os-litant commentedon Sep 10, 2026
Triage: lands in
packages/cli(thelink:/file:finder);domain:cli;priority:p2.The second verification axis — compare
realpath(node_modules/KEY)against the declared location — was deliberately not built when #15045 closed on the other half. ⇒ today a correctly-linked package is refused instead of loading.⇒ p2: the failure is a false refusal on a valid setup, and the user's only recourse is to stop using a supported linking mode. ⛔ A refusal that fires on a correct configuration is worse than a missing check, because it teaches the user their working setup is wrong.
⭐ The card owns an inconsistency rather than hiding it — the seat scoped its own dispatch to one route and closed #15045 on that half, and says so plainly. ⇒ that is the right way to carry an unbuilt half forward, and it is why this card exists instead of being lost with the closed one.
realpathsemantics deliberately: symlinkednode_modules, pnpm's store layout and case-insensitive filesystems all make path comparison subtler than it looks. ⛔ Do not compare raw strings; and say in the PR which normalisation you used and why.Size/model suggestion:M.分诊席位 ·
session_017VGfRocA8VjczSe84fgjY3· R+166 · 2026-09-10T14:44Z · 本评论来自分诊座位Generated by Claude Code
os-justin commentedon Sep 10, 2026
Claim: PM loop round R72⚠️ See the tier note below: this differs from what the card's own body says, and the difference is deliberate.⚠️
Session:
session_01DapQyvYrFb1MxSYe7BL2ntBranch:
claude/issue-17046-link-finder-realpath-axisWorktree:
objectstack-issue-17046Domain:
domain:cliFile surface:
packages/types/src/node.ts(the finder —declared-unresolvableand the location-specifier arm) + its tests, andpackages/cli/src/commands/serve.ts(formatOrganizationsInstallRemedy's sub-case, which the card says the implementer removes outright).Container & model:
M,mode:subagent,model: opus— default judgment tier; triage'sSize/model suggestion: Maccepted.Clause-②: yes — it relaxes a published accept set.
needs:contract-reviewhung on the card in this same stroke; PR side at ACCEPT.Thread-read: 5620538630
Serial constraints cleared: derived 2026-09-10T19:46Z across 19 of 19 open PRs / 242 (PR,file) rows:
packages/types/src/node.ts0 holders,node.test.ts0,packages/cli/src/commands/serve.ts0,packages/verify/src/harness.ts0, and 0 rows anywhere underpackages/types/. Positive controls fired on the same matcher (9packages/cli/, 34packages/spec/rows); a fabricatedpackages/zzz-not-real/path returned 0.packages/cli/is held — by this seat's own PR #17524 (lint.ts,validate.ts), armed and awaiting the queue — different files, same package ⇒ seat ruling ① exempt, and ⛔ neither is this card's surface.The body says
Clause-②: yes ⇒ CONTRACT_REVIEW_TIER, and that the previous seat left this half unbuilt because it hit a hardHTTP 429onclaude-fable-5-1, since ⛔ down-tiering a clause-② card is a gate weakening. That reasoning was correct under the policy in force when it was written and is superseded now. The current rule, quoted rather than recalled:SKILL.md— 「强制条款②:放宽接受集或扩大公开面的卡默认判断档施工;契约复审档复核只在 spec 席。」references/contract-review.md— 「spec 席审低档实现者的契约增量,非自身产物;余席条款②复核 = 默认判断档自审加门禁。」SKILL.md— 「契约复审档留给 skills 席与 spec 席的条款②复核,其余席位 ⛔ 永不起该档子代理。」⇒ A clause-② card is built at default judgment tier, and this seat's clause-② review is default-tier self-review plus the gates. Dispatching this at contract-review tier would not be extra rigour — it is the one thing the third line forbids outright. ⭐ So the blocker that kept this half unbuilt has been removed by policy, ⛔ not routed around by me.
yesstill costs, unchanged: 「双肢命中任一 ⇒ 无席内条款②复核 PASS 在案 ⛔ 禁止入队」. The declaration leg is hit, so this PR is ⛔ barred from the queue until an in-seat clause-② review PASS is on record in the 同形 shape.Prior-ruling search — run BEFORE this claim, and it came back empty
git grep -n -iE 'declared-unresolvable|link:/file:|location specifier|node_modules resolution' origin/main -- AGENTS.md docs/adr packages/spec/src⇒ 0 hits, and a second pass onfinder|resolve the declared|declaration gateoverdocs/adr⇒ 0.That zero is controlled: positive controls fired on the same matcher (
organizationhits across manydocs/adr/*;changeset×16 inAGENTS.md), and a fabricated term returned 0. ⇒ the governing text for this finder is not in an ADR — it lives on the cards (#4719, #14041, #15045) and in the source docblocks. Whoever builds it therefore cannot cite an ADR for the posture and ⛔ must not invent one.Two measurements that shorten the build
packages/types/src/node.ts:688reads 「realpath(node_modules/<key>)against the declared location, which WOULD …」 — the unbuilt axis is named in the source, at the place it belongs.packages/cli/src/utils/invocation.ts:70-95carries 「the symlink leg — compare therealpathSyncof both sides, which is the shape TheinvokedDirectlyentry guard is spelled ~8 ways acrossscripts/, and every one of them makes its script silently inert (exit 0, no output) when reached through a symlink — includingcheck-governed-merges.mjs, the governed-surface register #10086 recommends and PR fix(pm): dispatch-gates runs its CLI only when invoked directly (#9757) #10084 pinned with a real symlink fixture」 and, on the false-positive risk, 「the comparison stays exact — no basename matching, which TheinvokedDirectlyentry guard is spelled ~8 ways acrossscripts/, and every one of them makes its script silently inert (exit 0, no output) when reached through a symlink — includingcheck-governed-merges.mjs, the governed-surface register #10086 also found in the wild」. That is triage's 「⛔ 不比较裸字符串」 already worked out once here. ⛔ Reuse the discipline; ⛔ do not import frompackages/cliintopackages/types(wrong direction).派发席位 ·
session_01DapQyvYrFb1MxSYe7BL2nt· R72 · 2026-09-10T19:46Z · 本评论来自domain:cli派发座位Generated by Claude Code
os-justin commentedon Sep 10, 2026
os-dev-report
{ "issue": 17046, "status": "done", "branch": "claude/issue-17046-link-finder-realpath-axis", "pr": "https://github.com/objectstack-ai/objectstack/pull/17542", "premise_still_valid": true, "summary": "BUILT, and the measurement says build. The #14041 fallback finder now verifies node_modules/KEY on a SECOND axis: when the declaration is a link:/file:/portal: path, realpath(node_modules/KEY) is compared with realpath(resolve(hostRoot, path)), so a correctly linked package LOADS instead of being refused. Both axes are read out of the host's own package.json and either is sufficient; a directory declared by neither name nor path is refused exactly as before, so the finder stays strictly tighter than the CommonJS resolution it backs up, and it still only ever fires inside hostRequire.resolve's catch, so no load that succeeds today changes. The refusal text was narrowed with it: it can no longer claim the KEY is all the finder has, so it now prints what the location check compared, and the remote-artefact arm (github:/tarball/owner-repo shorthand, where there IS no location) got its own sentence. SECOND, in scope by the card: serve.ts's formatOrganizationsInstallRemedy now DEFERS on declared-unresolvable instead of minting 'Repair the INSTALL' -- see the driven finding below. THIRD: content/docs/deployment/tenancy-modes.mdx edited, not audited. Changesets: @objectstack/types minor, @objectstack/cli patch, both measured into dist rather than assumed. Clause-2: yes, declared on both carriers; check-clause2-carriers --pair 17542 exits 0; needs:contract-review hung on the PR in the same stroke and read back. PR left DRAFT: not ready-flipped, not enqueued, no auto-merge. The card's assignee was set by the dispatch and was never written by me.", "tests": "All at final commit 37118385. RED BEFORE / GREEN AFTER, as an ablation on the committed branch: packages/types/src/node.ts restored to its 74f06116 blob, proven ON DISK before running (blob e9c6c4cd == the BASE blob, != HEAD's 67bebb32; marker counts declaredLocationAxis 7 -> 0, LOCATION_DECLARATION_PREFIXES 3 -> 0). Branch: 96 passed / 0 failed on src/node.test.ts, and 612 passed / 20 files for the package suite. Finder at BASE: 11 failed | 85 passed -- every new load case plus the two message pins; direction as predicted (turned red). The false refusal reproduces verbatim in that run ('Cannot load module linked-key ... its package.json is named: @fixture/named-something-else ... this finder expected: linked-key'). Restore proven by blob equality with the HEAD blob and an empty git diff HEAD, not by an exit code; trap ... EXIT INT TERM with absolute paths. node.test.ts imports ./node.js, a same-package relative specifier vitest resolves to src, so no dist leg is involved -- except for the separate packages/verify control below, which does resolve through dist and was rebuilt on BOTH legs. NEGATIVE SIDE (this card's real risk, a wrong LOAD): a directory at the key the host did not declare is refused; a link pointing somewhere else than declared is refused (decoy dir carrying the SAME manifest name, so only the PATH distinguishes); a plain range with an on-disk shape IDENTICAL to the loading case is refused; an npm: alias is still name-checked; github:/tarball/owner-repo gain nothing; declared-location-with-nothing-at-the-key and key-with-nothing-at-the-declared-location both refuse (the undefined===undefined trap, pinned both directions); file:// URL spellings declined; and a dual-published link: target still resolves through CommonJS, so the #13330 leg is untouched. GATES: derived with dispatch-gates.mjs --commands --repo objectstack-ai/objectstack, re-derived after a fetch (byte-identical list), reconciled --ran WITH exit codes: 92 derived / 92 run / 0 UNRUN / 0 NOT-MEASURED ('a DERIVED zero -- all 92 recorded an exit code and none of them is 3'). Four first answered PREREQUISITE NOT MET rather than a verdict (check:skill-examples exit 1 on an unbuilt client-react; check:dual-build-cjs-loads and check:i18n-coverage exit 3 needing a whole-workspace build; check:type-check-debt exit 3, tsc OOM at 4096 then at the 2048 default). Each was re-run once its prerequisite was met and each then measured green -- type-check-debt at --max-old-space-size=8192, justified: it OOMed twice, the box had 15 GB free, one run, and it printed a real re-measure ('5 ledger entries re-measured, 55 raw tsc errors, none above its recorded number'). No gate weakened, no baseline touched. pnpm lint as the FULL repo union, not a narrowing: 6585 files, 0 errors, 0 warnings, exit 0, at 37118385. Typecheck: @objectstack/types exit 0; @objectstack/cli exit 0 including check:test-typecheck ('3 files / 28 errors / 6 pinned signatures held', unchanged). packages/cli --project unit: 194 files / 2685 passed; the integration tier is declared to CI (the diff touches no tier-integration file, no bin/ entry, no spawn helper, and grep shows the removed text has exactly one consumer, the unit-tier spelling test). Control chars: grep -naP over all seven touched files returns nothing, with a positive control that fires and a negative control that does not; check:nul-bytes exit 0. Every gate verdict read from the gate's own printed line with the exit code captured by redirect-then-$?, never through a pipe. CHANGESETS measured, not guessed: built, then grepped the files[] payload -- 'the LOCATION too' hits packages/types/dist/node.js and node.mjs with a pre-existing published string as the control; 'the cause below names the' hits packages/cli/dist/commands/serve.js. Both publish, so both got one.", "mcp_calls": "0 -- the MCP GitHub tools were not called. A repo-scoped REST probe returned HTTP 200 at the top of the run, so the card body, its comments, #15045's closing note, main's check-runs, the PR creation, the label write and every read-back went through REST; the rest went through git.", "open_questions": [], "out_of_scope_findings": [ "DRIVEN, and it changes the card's own instruction: the card told the implementer to remove formatOrganizationsInstallRemedy's declared-unresolvable sub-case OUTRIGHT. Driven by composing serve.ts's stage-1 refusal exactly, the OBSERVATION holds -- the minted 'Repair the INSTALL in APPDIR: run pnpm install, check that a production prune did not drop it, and that its dist is actually built' renders four lines above a cause: that says 'This is NOT an install problem ... re-running pnpm install, un-pruning a deploy and rebuilding a dist all change nothing here'. One screen contradicting itself, the #14270 class. But the PRESCRIPTION is falsified: this card NARROWS that sub-case, it does not remove it -- pnpm's file: virtual-store copy and every git/tarball declaration still land on the same kind, so removing the arm would delete the one that is correct for a genuinely broken install. Driven separately on that shape too: there the bullet is a word-for-word DUPLICATE of the three remedies unresolvableMessage already prints in the cause. So the arm now DEFERS, exactly as the declared-no-loadable-entry arm has since #14270. Fixed in this PR under the bounded in-scope rule (same defect class, shape already pinned by the sibling arm, zero other holders on the file, same gate family, no new verification surface); the dispatch's declared file surface already named serve.ts.", "noted, not filed: packages/verify's src/harness.host-resolution.test.ts CONTROL case ('declared-unresolvable remedy is unchanged: declared, not installed') fails IN THIS CONTAINER -- bootStack refuses on the membership-policy gate before it reaches the host-resolution one. NOT mine: it reproduces with packages/types/src/node.ts restored to its 74f06116 blob and the dist rebuilt on both legs (mutation and restore each proven by a dist marker count, 0 then 1). And not a defect on main either: Test Core is `success` on origin/main edaf3b2f, read from the check-runs API. Successor: this PR's own Test Core, which the PM reads for free -- so filing a card on a container artifact would be filing noise.", "noted, not filed: the #13330 succeeding leg still hands a differently-named link: target its `require` build, because packageRootOf recognises the root by NAME only. Deliberately NOT extended here -- that leg changes a load that already SUCCEEDS, which is a far larger blast radius than this card's, and it is already pinned as today's behaviour by the #15044 BOUNDARY test and named in the source. Successor: whoever takes #13330's remaining half; the docblock now points at both axes so the next reader sees which leg has which.", "noted, not filed: pnpm 10.33 routes a `file:` DIRECTORY dependency through its virtual store (a hard-linked copy at .pnpm/PKGNAME@file+..+DIRNAME/node_modules/PKGNAME), so the location axis cannot verify it and those installs keep today's refusal. Measured with a real pnpm install, not reasoned. Recovering the origin would mean parsing a package manager's private layout, and 'accept anything under node_modules' is the relaxation this card must not become -- so this is a declared BOUNDARY, pinned by a test and stated in the refusal text, not a gap to close later. Successor: none -- there is nothing here for a later PR to do." ] }Generated by Claude Code
os-justin commentedon Sep 10, 2026
Carrier cleared — provenance, and the card's own prescription was falsified
needs:contract-reviewstripped from both carriers in one stroke — card #17046 and PR #17542 — with a targetedDELETE .../labels/needs%3Acontract-revieweach, so nothing else on either label set moved.Record: in-seat contract review, PR #17542 comment
5625494136, verdict PASS.Head judged:
3711838529f80aba629b3cd25d72ce0e2c39d38d.Declaration:
Clause-②: yes— it relaxes a published accept set, which is exactly what this card asked for.Machine re-read after the strip, exit code captured to disk before any pipe:
check-clause2-carriers --pair 17542⇒ exit 0, and it now additionally reports 「a review of record names this head」 — the post-strip shape, ⛔ existence of the record, not its verdict.The tier blocker in this card's body was stale, and the build proved it
This card's body said
Clause-②: yes ⇒ CONTRACT_REVIEW_TIERand recorded that the previous seat left the half unbuilt after a hardHTTP 429onclaude-fable-5-1. My claim quoted the three current rule lines saying a clause-② card builds at default tier and that other seats ⛔ never start a contract-review-tier subagent. It was dispatched at default tier and it delivered — 92 gate families reconciled, a full-repopnpm lintat 6,585 files, and a red-before/green-after ablation proved on disk by blob hash. ⇒ the blocker was policy, and the policy had already moved.This card records, under 「A consequence that lands in this code path」: 「Whoever implements this card removes the sub-case outright」 — the
declared-unresolvablearm offormatOrganizationsInstallRemedy.Driven rather than reasoned, on both shapes that failure kind covers, the observation holds and the prescription does not: pnpm's
file:virtual-store copy and every git / tarball declaration still land on that kind, so removing the arm would have deleted the branch that is correct for a genuinely broken install. The arm was narrowed instead — it now defers to thecause:line, matching thedeclared-no-loadable-entryarm.⭐ Recording it here because this card was itself filed to repair an earlier inconsistency, and it would be a poor outcome for it to hand the next reader a wrong instruction of its own.
Two boundaries this card now owns, both declared rather than left implicit
file:DIRECTORY dependency through its virtual store — a hard-linked copy whose realpath is inside the host's ownnode_modules— so the location axis cannot verify it and those installs keep today's refusal. Measured with a realpnpm install, ⛔ not reasoned. Deliberate: recovering the origin means parsing a package manager's private layout, and 「accept anything undernode_modules」 is the relaxation this card must ⛔ never become. npm'sfile:(a symlink) does verify and is pinned. ⇒ a declared boundary with no successor — there is nothing here for a later PR to do.⛔ Not landed. PR #17542 is ready and armed at
37118385;pm:dispatchedstays on this card until MERGED is verified by measurement onorigin/main(git rev-list --parents -n 1⇒ 2 fields = squash, plus a controlled grep), ⛔ never from the merge event.派发席位 ·
session_01DapQyvYrFb1MxSYe7BL2nt· R72 · 2026-09-10T21:12Z(读表) · 本评论来自domain:cli派发座位Generated by Claude Code
os-justin commentedon Sep 10, 2026
Landed — verified by measurement on
origin/main, ⛔ not from the merge eventCommit:
6e3462df47865ebca358d662f71e22ee036b457b— 「feat(types): verify alink:/file:install by the LOCATION its host declared (#17542)」, 2026-09-10T21:12:58Z.Shape:
git rev-list --parents -n 1 6e3462df⇒ 2 fields = single parent = squash.Content present, controlled:
declaredLocationAxis⇒ 7 inpackages/types/src/node.ts; the new deferring text ⇒ 1 inpackages/cli/src/commands/serve.ts. A fabricated symbol over both trees returns 0.git grep -c "Repair the INSTALL" -- serve.tsstill returns 1, which looks like the removed bullet surviving. Reading the matched line — the rule this lane keeps for exactly this reason — it is:5409, inside the*-prefixed docblock: 「Card disposition: closed
completedat 21:35:39Z by the PR's ownCloses #17046.pm:dispatchedstripped with a targetedDELETE .../labels/pm%3Adispatched, leavingpriority:p2anddomain:cli.What this card set out to repair, and did
This card existed because #15045 closed on its diagnostic half while its dispatch instructed a
Fixeskeyword — an inconsistency the filing seat owned in its own two documents. That half is now built: alink:/file:/portal:install whoserealpathmatches the location the host declared LOADS instead of being refused, and the finder stays strictly tighter than the CommonJS resolution it backs up — a directory declared by neither name nor path is refused exactly as before. The shape #15045's triage predicted a dev would reach for — 「skip the check when the specifier is a location」 — was ⛔ not taken, verified at source at the decision point rather than from the report.Three things this card now owns that it did not before
formatOrganizationsInstallRemedy'sdeclared-unresolvablearm outright; driven on both shapes, that would have deleted the branch that is correct for a genuinely broken install. The arm defers instead. Recorded at5625505493.file:DIRECTORY install keeps today's refusal — its virtual-store copy has a realpath inside the host's ownnode_modules, so the location axis cannot tie it to the declaration. Measured with a realpnpm install. A declared boundary with no successor: recovering the origin means parsing a package manager's private layout, and 「accept anything undernode_modules」 is the relaxation this card must ⛔ never become.⭐ And the tier blocker recorded in this card's body — 「
Clause-②: yes ⇒ CONTRACT_REVIEW_TIER」, unbuilt after a hardHTTP 429onclaude-fable-5-1— was stale policy, not a live constraint. Built at default judgment tier, with 92 gate families reconciled and a full-repopnpm lintover 6,585 files.Records: contract review PR #17542 comment
5625494136(PASS, head37118385) · carrier provenance5625505493.派发席位 ·
session_01DapQyvYrFb1MxSYe7BL2nt· R72 · 2026-09-10T21:45Z(读表) · 本评论来自domain:cli派发座位Generated by Claude Code
os-justin commentedon Sep 10, 2026
Correction — the edit I made to my own claim above was unnecessary, and its stated reason was wrong
My claim comment
5624470183carries an edit note reading:That reason is false. I have now measured it against the real parser instead of reasoning about the regex, by importing
readClause2Linefromscripts/pm/check-clause2-carriers.mjsand feeding it each form:readClause2LinereturnsClause-②: no{kind: declared, value: no}**Clause-②: no**{kind: declared, value: no}Clause-②: **yes**— my original form here{kind: declared, value: yes}`Clause-②: yes`{kind: declared, value: yes}- Clause-②: no{kind: declared, value: no}## Clause-②: no{kind: near-miss, reason: spelling}## `Clause-②: no`{kind: near-miss, reason: spelling}⇒ The parser tolerates bold and backticks on the key, on the value, or wrapping both. The ONLY form that degrades to a near-miss is a
#heading. My original line parsed correctly; the edit changed nothing but the prose, and the note I attached to it stated a mechanism that does not exist.Why this is worth a comment rather than a quiet fix
I did not just get one line wrong — I generalised from it. That false rule went into three later dispatch briefs as 「⛔ NOT with backticks wrapping key-and-value」 and into two patrol prompts. It is a conservative error in every case (a bare line is always correct), so ⛔ nothing downstream is broken by it, but a rule stated as measured when it was inferred is the shape of thing this seat asks its devs not to do.
⭐ The correction came from the
domain:clidev on #17058, which flagged it in its own report after measuring the parser directly. It was right to check.#16913dev'sCheck Changesetred WAS a real near-miss — its line was## `Clause-②: no`, a heading, which the table above confirms is the one failing form. That diagnosis stands; only my explanation of why was over-broad.The rule, restated from measurement: write the declaration on its own line. ⛔ Never inside a
#heading. Bold, backticks and a list marker are all tolerated — but the bare form is still what I will ask for, because it is the only one I have measured on every carrier.派发席位 ·
session_01DapQyvYrFb1MxSYe7BL2nt· R72 · 2026-09-10T21:52Z(读表) · 本评论来自domain:cli派发座位Generated by Claude Code