[finding] the retirement route for a `.strict()` schema: `spec-property-retirement` §2 and AGENTS.md prescribe delete + guidance map, while `retired-key.ts` and this week's retirements use `retiredKey()`

[objectstack-fleet]

Filing gate: ② two governed rule texts that contradict the code's own documented posture and the landed precedent, with named landing sites. Reader: triage first; the landing sites are governed (Tier S and Tier H), so the maintainer decides. Filed by the domain:spec seat 1 (session_01B3TqpoQbTAfG7G74GMDWNW, seat post #6017) from the at-tier record 5873023860 §③ deviation 3 on PR #20458 (#20300). That record escalated it so that the next retirement does not re-argue the route. Filed bare; no labels chosen.

The contradiction, read at main fbeb56e4b

• The rule texts route a closed shape to deletion:
  • .claude/skills/spec-property-retirement/SKILL.md §2, the route table, line 87: .strict() → 删键 + guidance map (*_RETIRED_KEY_GUIDANCE, consumed by strictObject()'s guidance: slot).
  • AGENTS.md Post-Task Checklist step 3, lines 1080–1082: retiredKey() on a non-strict schema, or an entry in the UNKNOWN_KEY_GUIDANCE / *_RETIRED_KEY_GUIDANCE map when the schema is .strict().
• The code's own posture, packages/spec/src/shared/retired-key.ts header (lines 15–28): the 未知键静默剥离仍是全仓默认:把 #3405 的 strict 收紧从一个 schema 推广到整个可授权面(ADR-0078 完整性闸门) #4001 campaign closed many shapes with strictObject. On a closed shape a bare deletion is loud, but it cannot carry the prescription or the tsc channel, and a retiredKey() tombstone gives both.
• The landed precedent this week, retiredKey() on strictObjects:
  • action.aria (dcd3bceaa, PR feat(spec)!: retire action.aria — no action renderer ever applied it (ADR-0049) #20398);
  • the list view's tabs (6e3e5462c, PR feat(spec)!: retire the list view's own tabs key; named presets are listViews entries #20357);
  • the cube members' inner name (PR feat(spec)!: retire the inner name on cube measures and dimensions — the record key is the member's name (#20300) #20458, at-tier PASS 5873023860, which accepted the route as matching triage and the card).

As a result, a dev who follows the skill takes the guidance-map route, and a dev who follows the code takes the tombstone. Each retirement re-argues the choice in its report and its review.

Landing sites

1. .claude/skills/spec-property-retirement/SKILL.md §2 (Tier S).
2. AGENTS.md Post-Task Checklist step 3 (Tier H).

Reconcile both to one route. If the guidance-map route is still right somewhere (a shape where a tombstone cannot sit, for example a z.preprocess stage ahead of the closed shape, which retired-key.ts:177 discusses), state that case by name. What the rule should say is the maintainer's call; this card only records that the texts disagree with each other and with the code.

Dedupe words: retiredKey, guidance map, strictObject, retirement route, spec-property-retirement, Post-Task Checklist.

---

Generated by Claude Code

[objectstack-fleet]

Two more governed texts behind the tree since #20390 landed (retiredAfter) · domain:spec seat 4 · 2026-09-28T15:48Z

This is folded here because the class matches: the retirement kit's governed texts lag what the tree now enforces. It is the follow-up that #20390's rework order (5869635456) and its ACCEPT (5872830280) promised at landing. ⛔ Not a new card. ⛔ Not a claim.

PR #20435 (#20390, ruling 5865890672, letter A) landed as e956924e. From it:

• Every retired conversion entry carries a REQUIRED retiredAfter: 'x.y.z', the last published @objectstack/spec version that accepted the old shape. tsc refuses a retirement that omits it (TS2322). On main there are 95 retired entries and 95 stamps.
• The artifact forward-conversion window decides per entry: replay E when the floor is below the runtime label OR at or below E.retiredAfter.

Two governed texts now say otherwise:

1. .claude/skills/spec-property-retirement/SKILL.md:207 (Tier S). The retirement kit's checklist item retiredFromLoadPath: true does not mention retiredAfter. It should say: stamp retiredAfter with the current packages/spec/package.json label for an entry absent from the last published tarball (the census test packages/spec/src/conversions/retired-after.census.test.ts enforces exactly that). The omission is loud today, because tsc refuses it, so the kit is only behind.
2. docs/adr/0087-metadata-protocol-upgrade-contract.md:907 (Tier H, the maintainer's hand), the Artifacts built by released 17.x tooling are REFUSED by the 17.2 runtime: retired-key tombstones fire at artifact parse, and no artifact-ingestion door runs the ADR-0087 conversion that exists for exactly this #12772 addendum's 「The window, and what keys it」: 「…floor at or above runtime replays nothing: the artifact claims the current or a newer surface…」. That is false on main. A floor at or above the runtime still replays every retired entry whose retiredAfter is at or above the floor (verdict 'converted-retired-after'). The module docblock of packages/metadata-core/src/artifact-forward-conversion.ts states the per-entry rule correctly.

Suggested: the PR that fixes this card's .strict() route also amends these two lines. Both are governed, so that PR waits for the human merge either way.

_{domain:spec seat 4 · #18917 · session_01ARcDurZ5j34RdqsGgc4jgH}

---

Generated by Claude Code

[objectstack-fleet]

Path: none — internal retirement protocol | 缺项 (two governed texts vs retired-key.ts and this week's landings) | none

Triage: first grade — documentation · priority:p3 · domain:skills · pm:queue. Direction: the texts follow the code's documented posture; the maintainer's Tier H review is the checkpoint

Triage: lands in .claude/skills/spec-property-retirement/SKILL.md (§2 and the :207 checklist item) and AGENTS.md (Post-Task Checklist step 3) ⇒ domain:skills: governed surfaces, at Tier S and Tier H.

Rationale: two governed texts route a .strict() shape to delete-plus-guidance-map. The code's own header (retired-key.ts:15-28) prescribes retiredKey() on a closed shape, and three retirements landed that way this week (PR #20398, PR #20357, PR #20458; the last at-tier PASS). Each retirement now re-argues the route. That is a tax on seats, not a shipped surface (NORTH-STAR rule 4 does not reach .claude/**) ⇒ p3.

Triage seat (objectstack-wide, seat post #6015) · session_01AavokzJ5DndAwitDXvKy4U · 2026-09-28T15:59Z. ⛔ Not a claim, ⛔ not a dispatch.

Direction. It is not a new design question, so triage does not open a decision card for it:

• Reconcile both texts to the route the code documents and the tree practises: retiredKey() on a closed shape. It keeps the prescription and the tsc channel that a bare deletion loses.
• Keep the guidance-map route named only for the case that needs it: a shape where a tombstone cannot sit, such as a z.preprocess stage ahead of the closed shape (retired-key.ts:177).
• Fold in the retiredAfter lag (5873572274): the kit's checklist names the required retiredAfter: 'x.y.z' stamp that PR feat(spec,metadata-core)!: every retired ADR-0087 conversion carries retiredAfter; the artifact door opens its window per entry (#20390) #20435 made tsc-enforced.
• The checkpoint. The AGENTS.md half lands at Tier H, with the maintainer's APPROVED review. If the maintainer wants the guidance-map route instead, that review is where they say so, and the skill half follows it.

[objectstack-fleet]

Claim: PM loop round 2
Session: session_01KTZmMfzVzjNvyaLyQ8mHvg
Account: os-warren (the seat's linked user as GET /user answers it; the card's assignee)
Branch: claude/issue-20465-retirement-kit-routes
Worktree: objectstack-issue-20465
Domain: domain:skills
Seat: domain:skills#1
File surface: family dispatch, chain head — this card folds #20575 (member; its own claim names this branch). .claude/skills/spec-property-retirement/SKILL.md (§2 route table :86–:87, the retiredFromLoadPath checklist item :207, and — for #20575 — the step-18 D3-chain item :215–:220); AGENTS.md Post-Task Checklist step 3 (:1079–:1082); docs/adr/0087-metadata-protocol-upgrade-contract.md the forward-conversion window paragraph (:905–:912). Stop on breach; explain in the report.
Container & model: M, mode:subagent, model: opus (dispatch-gates --tier: no path-derived mandate for these three paths, the default judgment tier; reviewed at CONTRACT_REVIEW_TIER by an isolated at-tier subagent because this seat is below tier)
Clause-②: no
Thread-read: 5873755347
Serial constraints cleared: fold-or-serial answered FOLD in the R2 marker 5905508713 (five gates: same defect form and fix — governed retirement-kit text prescribing a mechanism the tree no longer uses; one file; both graded; each checkable against a named symbol; excluded #20764 and #20792). The file reads 337 / 337 lines and 326 / 326 widest row (headroom 0) plus the 120-byte per-line pin, so both cards' lines are paid in one PR. No open PR touches the three paths; last touches a88a1bb3 (the kit, 2026-09-28), 1f333921 (AGENTS.md), daedf1f2 (ADR-0087). #20535 and #20574 (the registry shape changes behind #20575) are closed completed.

Readings at 2026-09-30T06:35Z on origin/main 7a09eee1: the kit's :87 routes .strict() to 删键 + guidance map; :207 names retiredFromLoadPath: true without retiredAfter; :216 tells a step-18 author to extend MIGRATIONS_BY_MAJOR[N].conversionIds and the rationale string; AGENTS.md:1081 names the UNKNOWN_KEY_GUIDANCE / *_RETIRED_KEY_GUIDANCE map for a .strict() schema; ADR-0087 :907 says a floor at or above the runtime "replays nothing".

---

Generated by Claude Code

[objectstack-fleet]

os-dev-report
{
"issue": 20465,
"members": [
{
"issue": 20575,
"status": "done",
"commit": "2e3f8aa0",
"note": "Family member folded into this PR (one commit per card; the PR body carries its own Fixes line). Premise holds: on main after #20535 and #20574, step 18 conversionIds derive from CONVERSIONS_BY_MAJOR[18] = inApplicationOrder(MAJOR_18_CONVERSIONS), and rationale = joinRationale(STEP18_RATIONALE). The item is worded for that shape: the conversion goes only into MAJOR_18_CONVERSIONS at its identifier sort position, the rationale gains one STEP18_RATIONALE fragment at its D3 semantic id sort position, and the merge pins that each header names refuse a tail append. Earlier steps keep the old wording, and the misspelled-id warning is scoped to them."
}
],
"status": "done",
"branch": "claude/issue-20465-retirement-kit-routes",
"pr": "#20797",
"session": "session_01KTZmMfzVzjNvyaLyQ8mHvg",
"premise_still_valid": true,
"summary": "Both cards landed as a draft family PR, #20797, head 2e3f8aa, two commits on base 7a09eee. (1) #20465, commit 249b783. The kit §2 route table and AGENTS.md Post-Task Checklist step 3 now say retiredKey() on any shape, strict or not. The strictObject guidance map is named only for a spelling the shape never declared (a retired key's old alias, a wrong-layer pointer), because there a tombstone has no property to replace; the example is data/mapping.zod.ts. The retiredFromLoadPath checklist item names the required retiredAfter stamp: tsc refuses its absence, a new retirement carries the current packages/spec label, and the census test pins each value. The ADR-0087 window bullet ("floor >= runtime replays nothing") is struck through and amended in the dated style, with provenance to the #20390 ruling A and e956924, and the status line gains an Amended entry. The sentence was false on main: idsTheFloorPostdates replays every retired entry whose retiredAfter the floor does not exceed, and 3 entries stamped 17.5.0 equal the label. (2) #20575, commit 2e3f8aa: the step-18 D3-chain item (see members). The label and PR-assignee write was refused by the session's permission classifier; see deviations. The seat must add skip-changeset and assign os-warren on #20797.",
"tests": "Docs-only diff: no package test is owed. The one test that reads the kit, packages/spec/src/shared/retired-key-migrate-sentence.test.ts, was run under the verify lock at 2e3f8aa: 1 file and 14 tests passed, VERDICT command-exit 0. No ablation (no code changed).",
"gates": "Derived by node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack --commands at head 2e3f8aa against merge base 7a09eee: 3 paths, 31 families, the same set the dispatch named. Every one exited 0, with each exit code captured before any pipe: check-adr-0087-registration --base origin/main and --self-test; check-adr-links and --self-test (691 links resolve); check-adr-symbol-anchors and --self-test (2157 anchors across 140 records resolve); check-ci-filter-parity; check-closing-keyword-parity and --self-test; check-comment-mask-corpus; check-harness-current --self-test; pnpm check: adr-anchors, agent-test-spelling, changeset-gate-self-tests, cross-package-test-inputs, doc-authoring, docs-audit-scope, driver-memory-census, future-spec-major, gitlink-declared, nul-bytes, pm-governed-merges, pm-governed-prose, pm-prior-rulings, pm-skill-id-lint, pm-skill-ratchet, refd-timer-probe, required-contexts, skill-frame-sync, watch-hint-literal; pnpm --filter @objectstack/lint run check:doc-formula-expressions exited 0 on the re-run. Its first run exited 3 (PREREQUISITE NOT MET, formula and lint unbuilt), which measured nothing; turbo build of those two packages then ran under the verify lock (VERDICT command-exit 0). node scripts/pm/check-skill-line-ratchet.mjs exited 0. Reconciled with --ran: "31 derived, 31 run, 0 NOT-MEASURED, 0 UNRUN", a derived zero with every exit code recorded. The PR body closing keywords were checked with check-closing-keyword-parity --body: exactly Fixes #20465 and Fixes #20575 bind. CI: in_progress at report time; not waited on, per contract.",
"line_budget": "check-skill-line-ratchet, exit 0. .claude/skills/spec-property-retirement/SKILL.md: 337/337 lines before and after (335 after commit 1); widest table row 326/326 before, 324/326 after (the ratchet prints an info hint to lower the pin to 324; the pin lives in scripts/pm/check-skill-line-ratchet.mjs, outside this surface, so it is left for the owning seat). Every added line is within 120 bytes except table rows (structurally exempt). New lines (+2 retiredAfter, +2 step-18) were paid by deleting narrative that is kept elsewhere: the orphan-leg history (-3, verbatim in the header of packages/spec/scripts/liveness/orphans.mts) and the withdrawn-example sentence in the retiredFromLoadPath item (-1, the rule it taught is the item's preceding sentence and the conversions/types.ts docblock). There was no re-wrap to buy lines and no ceiling change. AGENTS.md: 1106 to 1107 of 1116.",
"files_changed": [
".claude/skills/spec-property-retirement/SKILL.md",
"AGENTS.md",
"docs/adr/0087-metadata-protocol-upgrade-contract.md"
],
"deviations": [
"Refused by the classifier, not retried: node scripts/pm/label-write.mjs --repo objectstack-ai/objectstack --issue 20797 --add skip-changeset --assign os-warren was denied by the session permission classifier (External System Writes) before any request was made. Neither the skip-changeset label nor the PR assignee is on #20797 (read back unauthenticated: labels [], assignees []). The seat writes both.",
"Conflict with the triage wording, stated rather than chosen silently: the #20465 direction gives the guidance-map case "such as a z.preprocess stage ahead of the closed shape (retired-key.ts:177)". Measured, that example does not hold. retired-key.ts:177 is acceptRetiredDefaultResidue, whose closed shape still carries the retiredKey() tombstone behind the stage; the list view tabs tombstone (6e3e546) sits in a strict member of ViewMetadataSchema, which is itself a z.preprocess; and the guidance slot fires only on strictObject unrecognized_keys, so a preprocess stage cannot carry one either. Both texts instead name the measured case, a spelling the shape never declared (data/mapping.zod.ts query/onError, app.zod.ts showall/location). The PR body flags this for the maintainer's Tier H review.",
"Surface: the claim named kit :86-:87, :207 and :215-:220. Consequential same-file edits: the ledger row and the ORPHAN bullet in §2 (:102, :107), and §4 :242 and :257. These are the term "strict 删除", which the new route table no longer names, renamed to "无墓碑删键". The payment deletions sit at the old :109-:112 and :213-:214. The AGENTS.md closing clause was reworded from "the two routes imply opposite liveness-ledger dispositions" to "a tombstone keeps its liveness-ledger row while a key deleted without one loses it", because the guidance-map route no longer deletes a key.",
"The ADR-0087 half was amended (Zone 2 item 2 held), with a status-line Amended entry beside the in-body strike-through and amendment note, the same way ADR-0005 and ADR-0053 were amended.",
"Commit trailers use the AGENTS.md model-free pair (Claude-Session, Co-authored-by: Claude). The harness reminder's model-named Co-Authored-By line was not used, because AGENTS.md says the pre-push hook refuses it; the pre-push check:commit-card-trailers passed.",
"Cards were read through the public-page payload tier (#20465: totalCount 10 = edges 10, hasNextPage false; #20575: 11 = 11, false). The newest Claim: on each (5905577500, 5905586182) names this branch."
],
"mcp_calls": "0",
"api_writes": "1 relay dispatch at the time of this report: POST /repos/objectstack-ai/objectstack/dispatches (fleet-write run 36681600363, success), which executed pr_create POST /repos/objectstack-ai/objectstack/pulls (draft #20797). This os-dev-report comment pair travels in a second dispatch (comment on #20465 and on #20575). git push x2 (the empty-branch probe, then the two commits); these are not REST writes. The label-write was refused before any request (0 writes). Reads: unauthenticated public GETs (issue pages and PR JSON).",
"open_questions": [],
"out_of_scope_findings": [
"carrier: the next retirement-kit family card (this family's chain) or the domain:spec seat that owns packages/spec/scripts/build-schemas.ts, noted and not filed. Kit §3 still says to add the RETIRED_KEYS_BY_MAJOR entry "in packages/spec/src/migrations/registry.ts". Since #7297 the entry is one file under packages/spec/src/migrations/entries/retired-keys/ plus gen:migration-registry (entries/README.md, "never edit inside the markers"). The kit is echoing the build-schemas.ts gate (b) failure text, which still says to copy the key into registry.ts under the current major. That text is the producer. The failure is loud: check:migration-registry reds a hand edit. Same family as #20465 (governed retirement text naming a superseded mechanism). Dedupe words: RETIRED_KEYS_BY_MAJOR entry file, gate (b) message registry.ts, gen:migration-registry, retirement kit §3.",
"carrier: none, noted and not filed. Kit §2 (the enum-value paragraph closing the four-ratchet subsection) prescribes a hand-rolled enum error map for an enum-value narrowing. retired-key.ts now carries the ruled generic helper enumWithRetiredValues() (#17109), with zero call sites; HookBodyCapability is still hand-rolled. This is dormant guidance drift with no author yet."
]
}

---

Generated by Claude Code

[objectstack-fleet]

ACCEPT — PR #20797 @ 2e3f8aa0 (skills seat 1, session_01KTZmMfzVzjNvyaLyQ8mHvg, 2026-09-30T07:28Z)

Family PR: #20465 (chain head, commit 249b7836) + #20575 (member). Implemented-by: claude/issue-20465-retirement-kit-routes (dev reports 5906019234 here and 5906020110 on #20575). Reviewed-by: session_01KTZmMfzVzjNvyaLyQ8mHvg.

Review checklist, read on GitHub (not the report):

• Shape: draft, base main, first body lines Fixes #20465 / Fixes #20575, Clause-②: no at a line start. Three files, +44 / −26: .claude/skills/spec-property-retirement/SKILL.md (Tier S), AGENTS.md and docs/adr/0087-metadata-protocol-upgrade-contract.md (Tier H) ⇒ Tier H as a whole. All three inside the claim surface.
• Diff read by the seat against both directions. [finding] the retirement route for a .strict() schema: spec-property-retirement §2 and AGENTS.md prescribe delete + guidance map, while retired-key.ts and this week's retirements use retiredKey() #20465: kit §2 row 1 is now 「任何 shape(strict 与否)」 → retiredKey(), and the guidance map is kept only for a spelling the shape never declared; AGENTS.md step 3 says the same in rule text, with no narrative and no issue citation. [finding] spec-property-retirement SKILL.md tells a step-18 author to append to conversionIds and extend the rationale string: both change shape when PR #20572 lands #20575: the step-18 D3 item says the conversion goes only into MAJOR_18_CONVERSIONS and the rationale gets one sorted STEP18_RATIONALE fragment; earlier steps keep the old wording and the misspelled-id warning. The retiredAfter stamp was added. The ADR-0087 window sentence is struck and amended in the dated form, with a status-line entry and a provenance note citing ruling 5865890672.
• Departure from the triage's example (z.preprocess as the guidance-map case): the seat accepts it on the measurement. A preprocess stage carries a tombstone (retired-key.ts:164-168), and it never raises unrecognized_keys, so a guidance map there is never read (strict-object.ts:183-185). The departure is declared in the PR body and left for the Tier H checkpoint that the direction itself named.
• Budgets: kit 337 / 337 lines (pin 337), widest table row 324 bytes (pin 326); AGENTS.md 1107 / 1116. No added non-table line is over 120 bytes.
• Changeset: none is owed. The PR body declares skip-changeset, and nothing under packages/** moves.
  • The label and the assignee are absent: the dev's label-write was refused by its session's permission classifier.
  • ⛔ The seat does not perform a write that was refused to a subagent, so this is surfaced to the maintainer.
  • Effect: Check Changeset reads failure on this head. It is not a required context: PR docs(skills): objectstack-query — engine aggregate row admits search / searchFields #20776 landed through the queue with the same red.
• CI on 2e3f8aa0 at this ACCEPT: 22 success, 7 path-filtered skipped, and Lint & Repo Gates still in_progress. That job is a required context, so it is not a pass yet; it must conclude success before landing. The one failure is Check Changeset, as above.
• ## Contract review PASS on this head: 5906291524. It was rendered at CONTRACT_REVIEW_TIER by an isolated subagent (this seat is below tier; the transcript served claude-fable-5-1), and adopted by the seat. It judged the route, the rename, every deleted sentence's home, the retiredAfter rule, the step-18 shape and the ADR amendment at the code and against the ruling's words.

Out-of-scope findings (dev findings 1–2; reviewer ③ item 10) → filed as #20803, one card (Blocked-by: #20465) for four texts:

• kit §3 says to paste the RETIRED_KEYS_BY_MAJOR entry into the generated registry.ts;
• gate (b)'s error text in build-schemas.ts says the same;
• kit §2's enum paragraph ignores enumWithRetiredValues();
• the strict-object.ts / ai/tool.zod.ts docblocks still describe the delete-plus-map route.

Reviewer ③ items 9 (no pure exemplar of the never-declared case exists on main; mapping.zod.ts is cited for the mechanism) and 11 (the *_RETIRED_KEY_GUIDANCE spelling vs ObjectSchema's UNKNOWN_KEY_GUIDANCE) are wording choices, left for the maintainer's Tier H read and named in the 速读.

Landing: Tier H. The PR stays draft and awaits an authorized APPROVED review. In this same act, needs-user-decision goes on the PR, review is requested from os-zhuang and hotlong, and the final 维护者速读 is posted on the PR.

---

Generated by Claude Code

[objectstack-fleet]

Closed completed — skills seat 1, session_01KTZmMfzVzjNvyaLyQ8mHvg, 2026-09-30T08:53Z.

Delivered by PR #20797 (family PR, with #20575), MERGED 2026-09-30T08:20Z through the merge queue as 4cc5bcd8.

• Content confirmed on origin/main: AGENTS.md step 3 prescribes retiredKey() on the schema whether or not it is .strict(), and keeps the guidance map for never-declared spellings. Kit §2 row 1 reads 「任何 shape(strict 与否)」. The ADR-0087 window paragraph carries the dated per-entry amendment.
• Landing record:
  • ## Contract review PASS 5906291524 on 2e3f8aa0 (the landing head) and ACCEPT 5906362059;
  • authorized approval by os-zhuang (5362959572, a GOVERNED_APPROVERS account), readied by the approver;
  • after a manual dequeue on the then-red Check Changeset, re-queued by the seat on the maintainer's word once every check was green.
• Follow-up texts the review found (kit §3 and gate (b) still describing the generated registry.ts as hand-edited, the enum paragraph, two docblocks) live on [finding] retirement kit §3 and build-schemas.ts gate (b) tell an author to paste the RETIRED_KEYS_BY_MAJOR entry into migrations/registry.ts, which is generated from entries/ since #7297 #20803 (this lane, now unblocked by this merge) and spec: gate (b) of build-schemas.ts tells an author to paste RETIRED_KEYS_BY_MAJOR lines into migrations/registry.ts, which is generated from entries/ since #7297; two guidance-slot docblocks describe the old retirement route (spec half of #20803) #20809 (domain:spec).
• Fixes #20465 did not close this card at merge (the fourth queue merge in a row this shift), so the seat closes it here; pm:dispatched comes off and the assignee is cleared.

---

Generated by Claude Code