Repository navigation
finding: two code comments still say 26 shared AUTHORING_RULES — the live registry has 38 #7491
Description
Activity
Routing repair + cadence re-check (@
origin/main1530870): appendeddomain:metadata— thefindinggrade is unchanged, no ownership taken.- Routing rationale (resolves the earlier leave-blank): every stale "26" site lives in
packages/metadata-protocol—src/protocol.ts:2632and:3011, plussrc/protocol.platform-schedule-org-gate.test.ts:545and:567— so the fix lands in one package ⇒domain:metadataby the lane table. The registry itself (packages/lint/src/authoring-rules.ts:371) needs no edit; de-numbering or re-counting happens at the comment sites only. - Premise re-verified, and the drift has widened: live registry counts 38 rows; the stale "26" mentions are now 4 code sites (the card filed 2). Still comment-only, nothing behavioral — HOLD stands; natural one-liner for a sweep pack in this lane.
本评论来自分诊座位 Routine（#5474 试点），不构成认领。
Generated by Claude Code
- Routing rationale (resolves the earlier leave-blank): every stale "26" site lives in
Findings cadence re-check @
origin/main4ed4160: HOLD —findingstands;domain:metadatakept as principal landing; one correction to the 05:11Z routing comment's rationale.Correction (public readout, publicly corrected): the 05:11Z comment's claim that "every stale 26 site lives in
packages/metadata-protocol" is wrong —packages/objectql/src/plugin.ts:76still reads "the 26 sharedAUTHORING_RULES" on today's main (it was in the card's own original table), pluspackages/objectql/src/plugin.authoring-channel.test.ts:21("all 26 shared"). The label outcome (domain:metadata) survives the correction — see inventory — but the single-package premise it rested on does not.Re-measured inventory (registry-count sites only, CHANGELOGs excluded as immutable history): live registry = 38 rows (
name:count inside theAUTHORING_RULESliteral atauthoring-rules.ts:371). Stale "26"-as-rule-count prose:protocol.ts:2698,:3077,runtime-authoring-gate.ts:133,:435,platform-schedule-org-gate.test.ts:545,:567,runtime-authoring-gate.test.ts:288(allmetadata-protocol), plus the twoobjectqlsites above — 9 sites, two packages, up from the card's 2 and the 05:11Z comment's 4.⚠️ Scope trap for whoever takes this: the ~12 test-file headers repeating "which is why all 26 …" refer to "all 26 of this package's (file, verb) pairs" (seeprotocol.save-flow-canonicalization.test.ts:25for the full sentence) — a different 26, not the rules registry. Do not blanket-edit the boilerplate; verify each site's referent.Disposition: still comment-only, nothing behavioral — HOLD. The fix prescription upgrades from "two one-liners" to a small de-numbering sweep (drop the count or phrase as "the shared
AUTHORING_RULES"), majority of sites inmetadata-protocol⇒domain:metadatastays; the twoobjectqllines are below splitting cost and ride the same PR (cross-lane prose, no contract face).本评论来自分诊座位 Routine（#5474 试点），不构成认领。
Generated by Claude Code
Findings triage round, 2026-08-11 (PM session, maintainer-directed: 「跑一轮集中定级」).
Graded: promote. Two comments teaching "26 shared AUTHORING_RULES" against a live registry of 38 is prose actively misinforming the next reader about gate coverage. S; re-anchor line numbers by grep before editing, per the card's own note.
finding→pm:queue(domain:metadata).
Generated by Claude Code
Deferred at batch selection (round 1,
domain:metadataseat) — recording the traps rather than just pushing it to a later round. Stayspm:queue, unassigned, dispatchable by whoever clears the two conditions below.Why deferred — two independent blockers, not one:
packages/metadata-protocol/src/protocol.tsis serialized and currently held. PR fix(metadata-protocol): listCommits no longer hides env-wide commit history (#7779) #7814 (listCommits misses env-wide sys_metadata_commit rows — same strict organization_id equality #7705 fixed one function away #7779) holds it forlistCommitsand is in the landing window. Behind it this seat's hot-file queue runs Four more strictorganization_idequalities left in protocol.ts — two measured (revertCommit / rollbackToPackageCommit), two unverified (duplicatePackage / reassignOrphanedMetadata) #7819 → A Studio-saved form authored withgroupsstill degrades on the REST public-form routes — the producer fold does not reach stored rows #7134 → [P2] 运行时授权门扩到 object 写入 + 全量 runtime-safe 规则快照(#4463 P1 之后) #4716. A two-line comment edit is not worth entering that queue out of order, andprotocol.tstook six merges in one day — so the:2632anchor in this card is already presumed stale (the card itself records it drifting from:2542). Re-anchor by grepping the comment text, not the line number.- It is a genuine two-domain landing.
packages/objectql/src/plugin.ts:76isdomain:engine-core; only themetadata-protocolline is this seat's. The triage seat deliberately left it unrouted for exactly this reason, and adomain:metadatalabel has since been applied — which does not grant this seat theobjectqlline. Either it goes as a cross-domain exception designated by the triage seat, or a lane already inside one of those two files takes its own line.
⚠️ The trap worth naming, because it is the card eating itself: the fix must not be "change 26 to 38". The count is 38 as measured at7e93c89d;AUTHORING_RULESgrows, and writing a fresh hand-maintained number reproduces the exact defect this card records — it just resets the clock. The card's own prescription is the right one: drop the number, or phrase it as "the sharedAUTHORING_RULES", or referenceAUTHORING_RULES.lengthwhere prose allows. Whoever implements this should not re-derive the count at all — needing to know it is the smell.
Generated by Claude Code
Deferred one round by the
domain:metadataseat — hot-file serialization, not priority. Recording so a later sweep does not read this as unpicked.Why deferred: one of the two comment lines is at
packages/metadata-protocol/src/protocol.ts:2632, and that file is held by #7774 (dispatched 04:03Z). ⛔ Never two agents inprotocol.tsat once — it took 6 merges in one day recently.Serial queue on
protocol.tsas of 04:0xZ: #7774 (in flight) → #7491 → #7736 → #6504 → #4716.Deliberately NOT bundled into #7774. #7774's dev is in this exact file and could take the line in passing, but that would widen a PR's declared surface to carry a second card — the seat treats "while you're in there" as how a reviewed surface stops matching its description. #7774 instead carries the repricing question: does landing it move this card's anchor? The answer comes back in that dev's report, which is the useful thing to know before dispatching this.
⚠️ The line anchors in this card's table will drift. They already have once (:2542→:2632, recorded in the body). Whoever takes this should locate both sites by their text — the26 shared \AUTHORING_RULES`` fragment — not by line number, and re-count the registry at dispatch time rather than trusting the 38 recorded here.The routing note still holds and is still the right call: two comment edits spanning
packages/objectql(domain:engine-core) andpackages/metadata-protocol(domain:metadata) is below contract-first splitting cost. This seat will take both lines in one PR.
Generated by Claude Code
Anchor correction — do not chase
protocol.ts:2632; use the comment text.PR #7931 (#7774) measured this card's anchor in passing and reported it. Recording it here so the next dispatch does not start from a dead line number:
- The card's body cites
packages/metadata-protocol/src/protocol.ts:2632. That line was already stale onorigin/mainbefore fix(metadata-protocol): key the /meta list merge on the i18n bundle discriminator (#7774) #7931 — theAUTHORING_RULEScomment sat at:2698. - fix(metadata-protocol): key the /meta list merge on the i18n bundle discriminator (#7774) #7931 pushes it further down again.
The substance is unchanged and still real: the comment claims 26 rules against 38 actually declared in
packages/lint/src/authoring-rules.ts. #7931 did not touch it.⇒ Anchor on the comment text (
git grep -n 'AUTHORING_RULES' packages/metadata-protocol/src/protocol.ts), never on a line number. This file took six merges in one day; line numbers in this lane have a shelf life measured in hours.Still queued behind #7774/#7931 on the
protocol.tsserial queue — this note is the deferral record, not a claim.
Generated by Claude Code
- The card's body cites
Claim —
domain:metadataPM seat (#6367). Sessionsession_01AmES43BMDg4bPrxTdi5q7t, branchclaude/issue-7491-authoring-rules-count. No competing claim.pm:queue→pm:dispatched. This card carried two public deferral notes; both are now discharged.⚠️ I am raising this lane's concurrent-editor cap from 2 to 3, and saying so rather than doing it quietlyUnder the maintainer's mutually-exclusive-region exemption I set a cap of 2 concurrent editors per hot file. That number was mine, not the protocol's — the exemption says declare regions at claim time · merge
mainbefore the PR · merge again after each sibling lands · let the queue arbitrate, and specifies no cap. I chose 2 conservatively on first adoption.packages/metadata-protocol/src/protocol.tscurrently holds #8027 (thesys_metadataoverlay path) and #7893 (thefieldwrite door). Measured this round, the rest of the queue is not an alternative: #6504 (IMetadataService) touches 14 files inmetadata-protocolandprotocol.tsitself; #7748 (the audit trail) touches 13; #7736, #7656 and #7529 likewise. The lane is not idle for lack of cards — it is idle for lack of disjoint cards.⇒ Under a standing instruction to accelerate, I am raising the cap to 3 and deliberately spending the third slot on the lowest-conflict card that exists, rather than on a large one.
⚠️ If this produces a conflict, it is my call and my cost — not the dev's. The mitigations are unchanged and are in the brief: anchor on function names, mergeorigin/mainbefore the PR, merge again after each sibling lands.Why this card is the right third occupant
It changes two code comments. That is very nearly the smallest possible conflict surface in an 11k-line file, and it is textually distant from both sibling regions.
The defect
Two comments state there are 26 shared
AUTHORING_RULES. The live registry inpackages/lint/src/authoring-rules.tsdeclares 38.⚠️ The anchor has moved twice — ⛔ do not trust any line number on this cardRecorded on this card at 09:0xZ and still true:
- the card's body cites
protocol.ts:2632. That was already stale before any of today's work — the comment sat at:2698onorigin/main. - fix(metadata-protocol): key the /meta list merge on the i18n bundle discriminator (#7774) #7931 (
7372d46) then moved it further down. - fix(metadata-protocol): an object extension reaches the by-name
/metaread, not just the list (#7556) #8015 (e3c8ed0) moved it again, ~30 minutes ago.
⇒ Anchor on the comment text (
git grep -n 'AUTHORING_RULES' packages/metadata-protocol/src/protocol.ts), never a line number. This file has taken many merges today and line numbers in this lane have a shelf life measured in hours.⚠️ Count the live registry yourself before writing a number. ⛔ Do not copy "38" from this card — it was measured hours ago and the registry is exactly the kind of thing another lane lands a change to. Whatever number you write must be one you counted, and the comment should ideally be phrased so it cannot go stale again.Brief carried inline in the dev's prompt.
Generated by Claude Code
- the card's body cites
Filed by the triage seat Routine (#5474 pilot) as the spin-off named in the 01:18Z round brief on #6015, so the verified fact does not die with PR #7483's body. Unassigned — recording, not claiming.
The drift
packages/lint/src/authoring-rules.tsexports the sharedAUTHORING_RULESregistry (declaration at:371onorigin/main@7e93c89d); it currently holds 38 rule entries (count ofname:fields in the array literal).Two code comments still describe the registry as having 26 entries:
packages/objectql/src/plugin.ts:76AUTHORING_RULESthatos validate/os lintrun)"packages/metadata-protocol/src/protocol.ts:2632AUTHORING_RULES"(The protocol.ts anchor was
:2632at verification time — an earlier readout recorded:2542; the file has drifted since.)Why it matters (and why it is only a finding)
Nobody hits this at runtime — both sites are prose inside comments. But both comments are load-bearing guidance about what the #4463 publish gate covers, and a hand-written count in two files has already gone stale once and will again. The mechanical fix is to drop the number or phrase it as "the shared
AUTHORING_RULES" (or referenceAUTHORING_RULES.lengthwhere prose permits).Routing note (deliberate leave-blank)
⛔ No
domain:*applied: the two comment edits land inpackages/objectql(domain:engine-core) andpackages/metadata-protocol(domain:metadata) — a two-domain landing for a two-line prose fix, below contract-first splitting cost. Left unrouted by the triage seat on purpose; named in the round brief. A lane picking this up alongside adjacent work in either file can take both lines in one PR.Generated by Claude Code