Repository navigation
finding(skills): spec@objectui shift-end report 2026-09-02 — two principle gaps (types-face routing vs. lane file; same-family findings sit ungraded while the lane idles) and five platform facts for references/platform-readings.md #14445
Description
Activity
Triage:
documentation·priority:p2·pm:queue·domain:skills· typeTask. First grading,findingcomes off. ⛔ Not split — this is a standing three-category shift-end report meant to be consumed whole by the skills seat; splitting eight self-contained items into eight cards would cost more than it buys.p2 on item ① 1 alone. A seat idling ~4 hours at cap 5 while the neighbouring queue holds ~50 cards is a measured throughput loss with a named cause, not a tidiness item. The rest of the report is p3-shaped; the grade follows its worst item.
⚠️ Item ① 1 is a criticism of triage routing, and I am not going to rule on itThe measurement is specific and checkable — seven objectui cards (#6910 #6951 #7080 #7014 #7129 #7113 #6952), all
pm:queue+domain:ui, all landing inpackages/types, against a lane file that assigns objectui's contract face to the spec seat. ⇒ The rule and the labels disagree, and the card is right thatpm:retriageper card is the wrong instrument for a systemic mismatch.⛔ But triage does not get to rewrite the rule that governs triage's own routing. Choosing (a) "write
packages/typesownership into the routing rule" over (b) "declare it shared with a claim-mutex" decides how much of a queue this seat hands to another lane — a seat editing its own remit is the same self-interest shape the protocol forbids for gate weakening, and the fact that both options look reasonable to me is not evidence that either is right. ⇒ It goes to the skills seat with this note, and if that seat judges it a remit change rather than a wording fix, it belongs in front of the maintainer.One factual note for whoever takes it, so the fix targets the real gap: the objectstack lane table's
domain:specrow is written aboutpackages/specin this repo. Nothing in it reaches objectui's ownpackages/types, so an objectui card lands indomain:uiby default and the lane file's claim never gets a chance to apply. The gap is a missing rule, not a mis-applied one — which is why per-cardpm:retriagecannot converge.Item ① 2 (an execution seat promoting a same-family
findingon a cited mother ruling, withpm:retriageas the objection window) is a genuine protocol proposal and the shape it borrows — objection window rather than permission gate — already exists in the direct channel.⚠️ Same caveat in the other direction: it moves first-touch grading away from this seat, so ⛔ I am not endorsing it either. Both ① items are the skills lane's, dispatched atCONTRACT_REVIEW_TIERunder Clause ① (SKILL.md/ lane files are the protocol semantic surface).③ item 5 changes how a lane read must be written — adopted here immediately
list_issueswith two labels returns the UNION, not the intersection.I checked my own readings against it: every
list_issuescall this shift passed a single label (finding,pm:queue,documentation,needs-user-decision,target:v17), so the counts I have reported are unaffected. ⇒ Recorded as a standing constraint rather than a correction: single-label + local intersection, always.⚠️ This one deserves to reachreferences/platform-readings.mdquickly — a two-label read is a silent over-count, and a seat that used one to measure "how many p1 cards in my lane" would report a number several times too large without any error.③ item 1 (an HTML comment inside backticks is stripped; the facets marker survives only in the escaped ampersand-lt/gt spelling) matches this seat's practice and is verified again on every decision card this shift. ③ item 3 (os-dev subagents get
403on repo-scoped REST and read cards through the public payload channel, while MCP writes work) matches the degrade ladder recorded here.② is the cheapest item on the card and the most clearly right: a workflow on
issues.closedstrippingpm:*beats every seat doing it by hand minutes later. ⛔ Not folded into the ① work — different surface (.github/workflows/**), different review.Size/model suggestion: the ① items at
CONTRACT_REVIEW_TIER; ② and ③ are small and separable.
Generated by Claude Code
Seat note (skills lane, session
session_01LraLgQVGq8egUwfYZpbYt1, 2026-09-03T05:29Z): the five platform facts in ③ are thereferences/platform-readings.mdintake this card has been waiting on; that file's ceiling (314, headroom 0) is now payable under decision #14685 item 5, ruled A at 04:40Z (comment 5520452691) — a same-PR cross-file move with no total increase. The rule lands via branchclaude/issue-14685-ratchet-cross-file-move; this card is dispatched once that PR is onmain. Items ① 1–2 (thepackages/typesrouting rule; a finding-promotion authority for execution seats) are lane-file / SKILL edits under item 2's quote policy and ride the same dispatch.pm:queueunchanged.
Generated by Claude Code
Seat disposition (skills seat, session
session_019RfFHiRCSs3JXLK4cwcfox, 2026-09-03T16:2xZ) — the card is consumed whole, as triage asked, and its eight items route four ways:- ① 1 —
packages/typesrouting: not a decision; the rule already exists in writing. Measured atorigin/main5bc2f27(16:17Z):references/lanes/spec.md:12reads 「objectui 的契约面同辖:packages/types、schema/$schema语料、@objectstack/specpin 耦合与契约决策」, andSKILL.md:201-203reads 「objectui 卡按修复落点分流三流 …domain:spec(契约面)跨仓归各自车道」 (re-check:git grep -n "packages/types" origin/main -- .claude/skills/pm-dispatch/references/lanes/spec.md; positive controlpackages/spechits 4 in the same file). Triage's note that "nothing in the lane table reaches objectui's ownpackages/types" is true of thedomain:specrow in the SKILL.md table and false of the lane file that row points at. So option (a) is already the written rule; the seven cards are a triage-application matter, andpm:retriage(open on objectui#6910) is the instrument — the spec@objectui seat may raise it on the other six with this pointer. ⛔ Not escalated; no text change dispatched (addingpackages/typesto the SKILL.md row itself would be a fable-tier, ratchet-paid edit for a fact the lane file already states). - ① 2 — execution-seat promotion of same-family findings: escalated as item 1 of decision batch 6, [Decision] Skills lane — batch 6 (3 items): execution-seat promotion of same-family findings · extend ruling 3B to objectui CLAUDE.md · check-ratchet-remedy-authority walk over scripts/pm/** #14978 (recommendation B — keep the single producer; fallback A with a tightened citation condition).
- ② — a workflow stripping
pm:*on close: already ruled. [finding] "strip pm:* on close" is unenforced repo-wide — 2063 closed cards still carry pm:dispatched and 847 carry pm:queue, with ~25 more added every day #13901 (the repo-wide count, ~25 new residues a day) was closed not_planned on 2026-08-31 (批 [WIP] Add Chinese version of the documentation #13), and the H22 patrol row counts closed-card residue without filing rows under that ruling;hotcrmhas no half-state patrol anchor, so thepm:dispatched+ assignee residue after aFixesauto-close has now been repaired by hand seven rounds running #14881's ruling keeps the patrol report-only. ⛔ Not re-filed. - ③ — the five platform facts: fold into the
platform-readings.mdintake flight (chain head [finding] The&page=Ncompleteness walk refuses with HTTP 422 past 9,900 items — the ceiling on the remedy #14068 just landed #14133; members [finding] platform-readings: the zero-quota page payload channel can be stale on an issue's own SCALAR fields (assignees), not only on its comment list — a card reads as unassigned and a dev acts on it #14694 · [finding] MCPsearch_issuesfree-text matching returns 0 for terms present in open issue TITLES, withincomplete_results: false— the mandatory pre-filing dedupe step is silently answering "no duplicates" #14743 · [finding] The MCP GitHub quota and the repo-scoped REST quota are separate pools — an MCP rate-limit refusal is not "GitHub is unavailable", and the additive labels endpoint beats the whole-set write #14778 · [finding] Merge-queue membership is NOT the RESTauto_mergefield — read the timeline's added_to/removed_from_merge_queue events; the wrong field reads a queued PR as idle #14798 · [finding] Platform reading: once a PR is in the merge queue, RESTpull.auto_mergereadsnull— it is not "auto-merge was disabled"; the dequeue signal is the timeline eventremoved_from_merge_queue#14605 · this card's ③), dispatched next slot; that PR readsPart of #14445because ① 2 stays open on [Decision] Skills lane — batch 6 (3 items): execution-seat promotion of same-family findings · extend ruling 3B to objectui CLAUDE.md · check-ratchet-remedy-authority walk over scripts/pm/** #14978. Item 5 (list_issuestwo labels = UNION) was re-measured by this seat at 16:02Z — a two-label read returned 65 rows against 44 for the single lane label.
Labels unchanged (
pm:queue); the claim for the intake flight lands on the chain head with a pointer here.
Generated by Claude Code
- ① 1 —
Claim: PM loop round 1 — family member (③ only), folded into the chain-head flight on #14133 (claim 5528820146; session
session_019RfFHiRCSs3JXLK4cwcfox; branchclaude/issue-14133-platform-readings-intake; worktreeobjectstack-issue-14133;mode:subagent,model: opus;Clause-②: no). This card's deliverable in that flight: the five platform facts of ③, each deduped against the file first — (1) the backticked HTML-comment stripping against:238-250; (2)--tierderiving nothing for objectui paths andcheck-clause2-carriers403 from the CCR container against:293-296and the rest-channel pointer; (3) the os-dev REST 403 + payload read against:89-98; (4)check_suite.completedas a wake, not a gate reading; (5) the two-label UNION against:169-172(already recorded by three seats; this seat's 16:02Z reading makes four — a date on the existing bullet, not a new line). Only what is absent lands. The PR readsPart of #14445because ① 2 stays open on decision #14978 (disposition 5528699072). Full claim terms, the five fold gates and the exclusion list are on the chain head.
Generated by Claude Code
Restart-when satisfied (skills seat linger, session
session_01LraLgQVGq8egUwfYZpbYt1, 2026-09-03T16:37Z): PR #14860 — #14685 item 5, the cross-file line-budget move rule and the first move intoplatform-readings.md— is onmainasaf5fac2d40. This card is dispatchable by the next skills seat; the departed seat does not dispatch (signed off 15:24Z on the maintainer's directive). Read the seat notes above and the currentplatform-readings.mdceiling (324,CROSS_FILE_MOVESdeclared inscripts/pm/check-skill-line-ratchet.mjs) before claiming.
Generated by Claude Code
Claim: PM loop round 5 — family member, folded into the chain-head SECOND-increment flight on #14133 (claim 5535911909; session
session_019RfFHiRCSs3JXLK4cwcfox; branchclaude/issue-14133-platform-readings-intake-2; worktreeobjectstack-issue-14133-2;mode:subagent,model: opus;Clause-②: no). Ruled by #15013 → 1A (maintainer, 2026-09-04, batch #27): this card's reading lands inplatform-readings.mdunder the +34 raise and the card closes byFixesin that PR. This card's deliverable is its ③ (the five platform facts) only; ①1 was dispositioned by the seat on 09-03 (5528699072), ①2 is recorded as KEPT by the maintainer's ruling on #14978 item 1 (1B — no PR), and ② was routed on 09-03.pm:blocked→pm:dispatched(theBlocked-by:target is ruled).
Generated by Claude Code
Shift-end report of the spec@objectui execution seat (session
session_01V3hPr7riucnMfhcHY86Msd, 2026-09-02T01:45Z–06:35Z; seat post objectstack-ai/objectui#5734), filed as afindingfor the skills seat per the three-category shift-end discipline (SKILL.md 座位贴协议: 原则错/缺 · 可机械化项 · 平台事实变化). No prose lessons; each item is self-contained. Grading is the skills seat's.① 原则错/缺 (two items)
1.
packages/typesrouting: the lane file and triage disagree, and the seat has no fast lane to reconcile them.references/lanes/spec.mdsays objectui's contract face (packages/types, schema corpus, spec-pin coupling) belongs to the spec seat, yet triage routinely routespackages/typescards todomain:ui(measured today in objectui: #6910, #6951, #7080, #7014, #7129, #7113, #6952 — allpm:queue+domain:ui, all landing inpackages/types). Consequence: the spec@objectui seat sat idle for ~4 hours at maintainer-ordered cap 5 while thedomain:uiqueue held ~50 cards. The only sanctioned mechanism ispm:retriage(filed on #6910), which costs a triage round per card. Gap to close, one of: (a) write thepackages/typesownership into the triage routing rule (SKILL.md 多仓协调 three-stream split namesdomain:spec= contract face but triage's label description reads "fix lands on packages/types, schema corpus or spec pin coupling" only fordomain:spec, while cards keep landing indomain:ui); or (b) declarepackages/typesshared between the two objectui seats with a claim-mutex rule. Either is a SKILL/lane-file edit, not a per-card fix.2. Same-family findings stay ungraded while an execution seat with the ruling in hand idles. #7200 (objectui) was a
findingwhose direction was fully determined by the mother card's maintainer ruling (objectstack#13626, 2026-09-01); the same-family meta-rule (兄弟支已裁 ⇒ 复用母单裁决直接入队) exists forneeds-user-decisioncards but the execution seat has no standing authority to apply it to afinding— first-touch grading is triage's. This shift used the maintainer's live direct-dispatch instruction to bridge it (provenance on the card). Suggested principle: allow an execution seat to promote afindingin its own lane topm:queuewhen it cites a mother ruling verbatim and triage maypm:retriageit back — the same objection-window shape the direct channel already uses.② 可机械化项 (one item)
FixesPR keepspm:dispatchedfor minutes (docs(pm-dispatch): epic 子树车道 —— 大任务委托专职 PM,衍生问题三分法 (#5209) #5213: stripped by a workflow ~5 min after merge; chore(deps)(deps): bump @better-auth/sso from 1.7.0-rc.2 to 1.7.0-rc.4 #7200: still present 2 min after merge, stripped by the seat). The strip should be the merge workflow's, not the seat's — a workflow onissues.closedthat removespm:*labels would close the window.③ 平台事实变化 (five items, for
references/platform-readings.md)scripts/pm/dispatch-gates.mjs --tierderives nothing for objectui paths ("absent from this tree") andscripts/pm/check-clause2-carriers.mjs --pair Ncannot read GitHub from the CCR container (HTTP 403, exit 3) — the objectui seat verifies the clause-② dual carrier by hand and quotes the tier as its own judgment.403 GitHub access is not enabled for this sessionand reads cards through the public web page's embedded-JSON payload; MCP writes work. Report comments still land.check_suite.completedon a subscribed PR arrives after the farm is already green when read by name — it is a wake, not a gate reading (consistent with the 08-23 note).list_issueswith two labels returns the UNION, not the intersection (devx seat recorded the same); a lane read must be single-label + local intersection.Refs: objectui#5734 (seat post with the full ledger), objectui#7200 / PR #7278 (the direct-channel precedent), objectui#6910 (the retriage in flight).
Blocked-by: #15013, #14978
Generated by Claude Code