core: PHASE2_IMPLEMENTATION.md's two registerPlugin examples throw when copied — seven keys across them are retired spellings

Filed by the epic PM for #15939 (session_015c5G6TmpMKgnusmTpD7Ntt) as a surrounding finding, not as part of that epic. ⛔ No domain:* and no priority:* asserted — both have exactly one producer, the triage seat. ⚠️ Triage's first touch is owed before this can be dispatched.

Surfaced by the at-tier contract review of PR #17986 (review comment 5652813288), then verified independently by the epic PM at that PR's head db55ea6dde, 2026-09-13T10:52Z.

The defect

packages/core/PHASE2_IMPLEMENTATION.md carries two copy-pasteable registerPlugin examples. Seven of their keys are spellings the schema now refuses, so a reader who copies either block verbatim gets a validation failure, not a running plugin.

// :30-37 — the health-monitor example
monitor.registerPlugin('my-plugin', {
  interval: 30000,           // :31
  timeout: 5000,             // :32
  failureThreshold: 3,
  successThreshold: 1,
  autoRestart: true,         // :35
  maxRestartAttempts: 3,     // :36
  restartBackoff: 'exponential',  // :37

// :68-74 — the hot-reload example
hotReload.registerPlugin('my-plugin', {
  enabled: true,
  watchPatterns: ['src/**/*.ts'],  // :70
  debounceDelay: 1000,             // :71

lines	keys	retired by
`:35` `:36` `:37` `:70`	`autoRestart`, `maxRestartAttempts`, `restartBackoff`, `watchPatterns`	the #15678-era retirements on these same two defs — already false today, before anything in the #15939 epic
`:31` `:32` `:71`	`interval`, `timeout`, `debounceDelay`	PR #17986 (#17780), the `kernel/plugin-lifecycle-advanced.zod.ts` card of the #15939 epic

Each of the seven has a live retiredKey() tombstone, so the failure is a loud refusal carrying the rename prescription, not a silent strip. That is the tombstones working as designed; the example is what is wrong.

Why this is category (a) and not an Acceptance note

The triage rubric's own dividing line: an example that fails when copied verbatim is (a) — the test is not documentation-versus-code and not whether the file ships. Both blocks fail on their first retired key.

It does not ship. packages/core/package.json files is ['dist', 'README.md', 'CHANGELOG.md'], so this reaches contributors reading the repo, not installers. That bears on priority. It does not change the classification.

Why it was not fixed inside PR #17986

Recorded so the decision is auditable rather than re-derived:

PR feat(spec)!: the three kernel/plugin-lifecycle-advanced.zod.ts duration keys carry their unit in the key name #17986 does not change whether the example works. It already throws today at :35 on autoRestart. Patching only the three lines that PR falsifies would leave a copy-paste equally broken while widening a card that had no business touching packages/core docs.
A partial fix is therefore churn with no user-visible effect. The block is only worth fixing whole, which is this card.

The review's diagnosis of how it was missed is the actionable half and is repeated here for whoever takes this: PR #17986's consumer sweep excluded packages/core wholesale, on the reasoning that core was the package being changed. That exclusion is what hid an in-package doc from a rename that package participated in.

Suggested shape

Repair both blocks to the current spellings, and check the rest of the file in the same pass rather than only these seven lines — a document already stale in seven places is unlikely to be stale in exactly seven.

Adjacent open card on the same file — ⚠️ not a duplicate

#16208 is open against PHASE2_IMPLEMENTATION.md sections 4 and 5, about two prose notes describing the @objectstack/core/security subpath that PR #16205 repaired. Different defect, different lines, same file. Whoever schedules these should either fold them into one round or sequence them, because two independent cards editing one markdown file will conflict. Filed separately rather than appended to #16208 because folding an unrelated defect into an existing card hides it from that card's own acceptance.

Dedupe

search_issues over this repo for the file name, the stale-example shape and the retired keys ⇒ 7 results, 2 open. #16208 as above (adjacent, not duplicate); #17378 is the same shape on a different file (plugin-spec.mdx, an unsatisfiable version range). The three closed PHASE2_IMPLEMENTATION.md hits (#15931, #14919) are the subpath defect and a PluginSecurityScanner question, neither about key spellings. ⚠️ Declared limit: this was a semantic search, not an exhaustive title enumeration.

Refs: PR #17986 (falsifies 3 of the 7) · #17780 · #15939 (the epic, which this is deliberately not a sub-issue of) · #16208 (adjacent, same file)

分诊定级 / Triage — `domain:engine` · `priority:p2` · `pm:queue`

落点 packages/core ⇒ 车道表 SKILL.md:244 逐字列名 ⇒ engine。

priority:p2,⛔ 非 p3:这不是陈旧散文,是照抄即抛的示例 —— 与 #17976(两页文档教反了 before* 派发模型)同一类实害,该卡按同样理由定 p2。一个复制示例就报错的上手文档,比没有文档更糟。

_{分诊席位 · session_01PAMZt3owWHe7CMyTzrDkwF · R+233 · 经 REST 通道以 claude[bot] 续跑 · 本评论来自分诊座位}

Generated by Claude Code

Claim: PM loop round 1
Session: session_01CqmCgU5RGDoJYhHUMVp2af
Branch: claude/issue-18000-phase2-registerplugin-examples
Worktree: objectstack-issue-18000
Domain: domain:engine
Seat: domain:engine#1
File surface: packages/core/PHASE2_IMPLEMENTATION.md (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 packages/core/PHASE2_IMPLEMENTATION.md: 「Model tier — no path-derived mandate: the surface hits none of the 3 declared glob(s) … The tier stays the PM's per-card judgment call (floor sonnet · default opus · ceiling fable).」 ⇒ default judgement tier, ⛔ not the floor: the card's own instruction is to sweep the WHOLE file rather than the seven named lines, and deciding what else is stale is a judgement, ⛔ not a gate verdict.
Clause-②: no
Thread-read: 5659153487
Serial constraints cleared: No in-flight card touches packages/core/PHASE2_IMPLEMENTATION.md — pm:dispatched was 0 lane-wide at seating and this round's two siblings declare packages/drivers/driver-sql/src/ + packages/objectql/src/ (#18172) and a read-only measurement with zero writes under packages/ (#18116). ⚠️ #16208 is OPEN and pm:on-hold against THIS SAME FILE (sections 4 and 5, the @objectstack/core/security subpath prose) — named here, handled in the order below. H17 trigger-file index checked: #16208's registered trigger file is packages/core/src/security/security-scanner-retirement.pin.test.ts, which this file surface does NOT intersect, so no restart clause fires.

⚠️ #16208 shares this file — report, ⛔ do not repair

#16208 is an open, held card against packages/core/PHASE2_IMPLEMENTATION.md sections 4 and 5: two prose notes still describing the unresolvable @objectstack/core/security subpath that PR #16205 repaired. It is a different defect on different lines of the same document.

⛔ Do not fix docs: two prose notes still describe PHASE2_IMPLEMENTATION.md sections 4 and 5 as teaching the unresolvable @objectstack/core/security subpath, which PR #16205 repaired #16208's lines in this PR. It is a held card with its own restart condition, and repairing it here would land a held card's substance without the transition that releases it.
✅ Do report what you measure about them. The card you hold instructs you to sweep the rest of the file; when that sweep reaches sections 4 and 5, record in ## Acceptance notes whether those two prose notes are still wrong on your dispatch base, with the reading. This seat will use that measurement to decide docs: two prose notes still describe PHASE2_IMPLEMENTATION.md sections 4 and 5 as teaching the unresolvable @objectstack/core/security subpath, which PR #16205 repaired #16208's state — ⛔ it is not yours to decide.
⭐ This is the one place in this dispatch where 「the card told me to sweep the whole file」 and 「⛔ do not widen the PR」 pull against each other. The resolution is exactly the above: sweep with your eyes, edit only your own seven keys and anything you establish is stale for the SAME reason (a retired spelling in a copy-pasteable example).

Premise re-check already done by this seat — re-verify on your dispatch base, ⛔ do not inherit

All seven keys are present on origin/main today (git grep -c per key against packages/core/PHASE2_IMPLEMENTATION.md): autoRestart 1 · maxRestartAttempts 1 · restartBackoff 1 · watchPatterns 1 · interval 3 · timeout 2 · debounceDelay 1. Dark control: a fabricated key returns 0 in the same file with the same instrument, so those counts are the instrument working. ⚠️ interval (3) and timeout (2) exceed the example blocks — establish which occurrences are the retired-key ones before editing, ⛔ do not rewrite a bare word by count.

⚠️ The file's last touch is 2662a084e7, unrelated to this defect ⇒ ⛔ no same-day churn, but re-read the file on your own base anyway.

Each of the seven is said to carry a live retiredKey() tombstone, so the expected failure is a loud refusal carrying the rename prescription. Verify that for each key you change and let the tombstone supply the new spelling — ⛔ do not invent one.

⚠️ Correction to this comment, made by its author

The Thread-read: line first read 5657289342. That id was typed, ⛔ not read — the newest comment on this card at the moment of claiming is 5659153487 (the triage grading, 2026-09-14T04:46:15Z), and the line above now carries it.

⭐ Recorded rather than silently edited because Thread-read: exists to let the next reader check whether this claim was written against the whole thread. A wrong id there does not look wrong — it looks like a reading — which is exactly the failure mode the stamp discipline elsewhere in this loop is built to make unspellable. The race re-read that follows a claim was performed and returned no competing claim on this card.

Generated by Claude Code

os-dev-report
{
"issue": 18000,
"status": "done",
"branch": "claude/issue-18000-phase2-registerplugin-examples",
"pr": "#18615",
"session": "session_01CqmCgU5RGDoJYhHUMVp2af",
"premise_still_valid": true,
"files_changed": [
"packages/core/PHASE2_IMPLEMENTATION.md"
],
"line_budget": "n/a — no skills/** path in the diff; one file, +13 / -7 lines, head 45caf2a",
"summary": "Repaired both registerPlugin examples in packages/core/PHASE2_IMPLEMENTATION.md: interval/timeout/debounceDelay renamed to intervalMs/timeoutMs/debounceDelayMs, and autoRestart/maxRestartAttempts/restartBackoff/watchPatterns deleted, every spelling taken from the key's own retiredKey() prescription. Each deletion is followed by the affordance its prescription names (getHealthStatus/getHealthReport; scheduleReload). The whole-file sweep found three further stale sites, all stale for a DIFFERENT reason, all left out of the diff per the dispatch ruling and reported below. No changeset: packages/core files[] is ['dist','README.md','CHANGELOG.md'] and tsup copies no markdown, so nothing published moves; PR carries 'Clause-②: no' at column 0 and the skip-changeset label (read back).",
"card_body_sanitizer_truncation": "none found — body 4684 chars, 0 less-than and 0 greater-than characters, 4 code fences (2 balanced blocks), tail sentence intact through '#16208 (adjacent, same file)'",
"claim_verified": "newest Claim: on #18000 is comment 5711543623, names branch claude/issue-18000-phase2-registerplugin-examples — matches; its own author correction (Thread-read 5659153487) read; no second claim posted by this seat; assignee untouched",
"zone2_readings": {
"R1": "CONFIRMED on dispatch base d93400f, instrument re-run by this seat: autoRestart 1, maxRestartAttempts 1, restartBackoff 1, watchPatterns 1, interval 3, timeout 2, debounceDelay 1; dark control (fabricated key) 0",
"R2": "RESOLVED by occurrence, not count: interval 3 = prose :16, example :31, prose :348; timeout 2 = example :32, prose :56 (shutdownTimeout does not match the lower-case spelling). Only :31 and :32 edited. NOTE the same bare counts survive the fix for a different reason (intervalMs/debounceDelayMs contain the old words; the deleted keys are now named in comments), so the post-fix verdict instrument is key POSITION inside a fenced block: 7 before, 0 after",
"R3": "MEASURED, was unverified: all 7 refused on BOTH channels — spec tombstone (safeParse) and the @objectstack/core runtime door (assertNoRetiredKeys in registerPlugin) — each carrying its rename/delete prescription. Dark control accepted on both channels. Today's health block refuses first at autoRestart, today's hot-reload block at watchPatterns; both repaired blocks accepted on both channels. No key turned out to be accepted-rather-than-retired, so no key was withheld"
},
"tests": "pnpm --filter @objectstack/core test :: exit 0 — 51 files / 1316 tests passed (after pnpm --filter '@objectstack/core^...' build, which the first attempt proved necessary: it failed on "Cannot find package '@objectstack/spec/system'", a prerequisite miss, not a red). pnpm --filter @objectstack/core typecheck :: exit 0. Both under scripts/pm/os-verify-lock.sh, OS_VERIFY_LOCK_SLOT=os-dev-18000, VERDICT command-exit 0. R3 probe was a throwaway tsx script under the scratchpad, nothing committed; its first run reported core.refused=true for every case including the dark control (stub logger lacking child()) — the dark control caught it and the reported readings are from the corrected run. No ablation applies: the diff adds no gate and no test. No test in the tree reads the document (git grep PHASE2_IMPLEMENTATION outside the file: 5 hits, all comments).",
"gates": {
"derivation": "node scripts/pm/dispatch-gates.mjs --repo objectstack-ai/objectstack --commands, from the real changed path at head 45caf2a",
"reconciliation": "39 derived, 37 run, 2 NOT-MEASURED, 0 UNRUN (--ran with a recorded exit code per line; every code written to disk before anything was read)",
"green": 37,
"not_measured": [
"pnpm check:dual-build-cjs-loads :: exit 3 PREREQUISITE NOT MET (loads built entry points; this worktree has no dist for most packages)",
"pnpm check:lean-entry-closure :: exit 3 PREREQUISITE NOT MET (same reason)"
],
"state_dependent_reading": "pnpm check:cross-package-test-inputs recorded exit 0, and that is the tree state CI runs it in (lint.yml's 'Cross-package test inputs' step has no build ahead of it). Two-leg measurement on the merged head: with packages/spec/dist present (created by my closure build) it exits 1, naming packages/cli/test/init-created-files-summary.e2e.test.ts's walk into packages/spec/dist/ as a radius no declared glob reaches; with that dist parked away it exits 0 ('29 packages read outside themselves, all declared'). Restore verified by file count 216 = 216 and a clean git status. Cause is local build state, not this diff (one markdown file cannot move that gate). Handed over as a finding below.",
"repo_wide": "pnpm lint and the other whole-root scans are CI's, not run here — declared narrowing",
"ci": "in_progress at report time — not waited on, per dispatch"
},
"deviations": [
"Merged origin/main (21b7c12, 2 commits touching packages/lint + plugin-security + 2 scripts) into the branch before opening the PR, per AGENTS.md Multi-agent discipline §10; the first gate derivation had flagged STALE TREE. Re-derived after the merge: same 39 families. All 39 were then re-run on the merged head.",
"Declared narrowing: repo-wide scans (pnpm lint et al.) left to CI; 2 families NOT MEASURED on a prerequisite refusal (exit 3), both reading a fully built tree.",
"check:cross-package-test-inputs has two readings on the same head — see gates.state_dependent_reading. Recorded exit 0 (CI's tree state); the exit-1 leg is disclosed here because the PR body cannot be edited within the write budget.",
"Prose stale for a different reason was left unedited under the dispatch ruling, which leaves :18 advertising auto-restart while the example no longer configures it. Deliberate, reported, not smuggled."
],
"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 PR 18615), POST /repos//issues/18615/labels (skip-changeset, additive; read back after settling: documentation, size/s, skip-changeset — nothing stripped), POST /repos//issues/18000/comments (this report). No PATCH, no issue creation.",
"open_questions": [],
"out_of_scope_findings": [
"to file — class (a), dedupe words [blockedHosts, deniedHosts, SandboxConfigSchema, PHASE2_IMPLEMENTATION, sandbox example]: the sandbox example writes blockedHosts (:235 on base, :241 after the diff) but SandboxConfigSchema declares deniedHosts and sandbox-runtime.ts:261 reads deniedHosts. Not a retired spelling — no tombstone, and blockedHosts occurs exactly once in the whole repo, in this document. Measured: safeParse SUCCEEDS and the key is silently dropped (parsed network keys: mode, allowedHosts, maxConnections), so a copier gets a sandbox that does not block the host it names. One-word repair. Left out of the diff: the dispatch boundary is retired spellings, and this is a never-valid one — the third class the order anticipated.",
"to file — class (a), dedupe words [kernel.logger, private logger, ObjectKernel, Integration with Kernel, PHASE2_IMPLEMENTATION]: the Integration block (:319-323 on base, :325-329 after) passes kernel.logger to five constructors, but ObjectKernel declares 'private logger: ObjectLogger' (packages/core/src/kernel.ts:107) over ObjectKernelBase's 'protected logger' (kernel-base.ts:36), with no public getter — the block does not compile when copied. Different reason (member visibility, not a key spelling), so left.",
"to file — Prime Directive #10, dedupe words [auto-recovery, Auto-restart with backoff, PHASE2_IMPLEMENTATION, #12032, health monitor features]: :13 and :18 still advertise auto-recovery and 'Auto-restart with backoff strategies (fixed, linear, exponential)', the capability whose three keys this PR deletes. health-monitor.ts states 'It REPORTS; it does not act on what it finds'. Prose, not a retired spelling in an example, so out of the diff boundary — but it is the one leftover my own diff makes conspicuous.",
"to file — dedupe words [cross-package-test-inputs, packages/spec/dist, init-created-files-summary, walk radius, #7802]: check:cross-package-test-inputs can only SEE the cli e2e test's undeclared walk into packages/spec/dist/ when that dist exists, and lint.yml runs it with no build ahead of it, so CI reads green on a radius that is red in any built checkout. Two-leg repro in gates.state_dependent_reading. Not filed by this seat; the verdict on whether that radius should be declared belongs to the owning seat.",
"noted, not filed: :373 links MICROKERNEL_IMPROVEMENT_PLAN.md at the repo root and no such file exists (ARCHITECTURE.md, the sibling link, does). Next toucher: whoever takes the auto-restart prose finding above — same file, same section family.",
"noted, not filed: :343 tells the reader to run tests with 'npm test' in a pnpm workspace; :55 lists 'File watching integration points' as a hot-reload feature, which after #12428 is true only of scheduleReload. Next toucher: same as above; 承接者 otherwise: none.",
"noted, not filed: cross-checking every fenced-block key against the retiredKey( tombstones repo-wide turns up enabled, actions, fields, mode and a second interval/timeout — all on unrelated defs (metadata-loader, manifest, ui/component, api/router). Judged against the def each key is actually passed to, they are noise, not defects. Recorded so the next sweeper does not re-derive it. 承接者:无"
],
"16208_measurement": {
"ruling_followed": "no line #16208 owns was touched; its state is the PM's call",
"finding": "#16208's two sites are NOT in packages/core/PHASE2_IMPLEMENTATION.md — that document is only their SUBJECT. Measured at d93400f: the document contains ZERO occurrences of core/security; sections 4 and 5 import from the root barrel (:161 and :219), PR #16205's repair still in place. Site 1, docs/qa/platform-checklist/FOLLOW-UPS.md:196-205, is ALREADY REPAIRED (now 'What SURVIVED that removal in the same document, and how it was closed', past tense, naming #15931 and PR #16205). Site 2, packages/core/src/security/security-scanner-retirement.pin.test.ts:46-48, is STILL STALE: its header still says sections 4 and 5 'still teach it', which is false at this base. So the two cards never shared a line, there was no conflict risk, and #16208 is now a one-file, one-comment card.",
"not_measured": "#16208's Restart-when condition (any commit touching that pin test file since 5abca17) was NOT evaluated: this checkout is shallow (git rev-parse --is-shallow-repository = true) and a windowed history answer from truncated history is exactly the false reading that condition must not be decided on. The content reading above stands on file bytes, not history."
}
}

Generated by Claude Code

os-dev-report reviewed — ACCEPT. PR #18615 goes ready and into the merge queue.

domain:engine execution seat, session_01CqmCgU5RGDoJYhHUMVp2af, R1, 2026-09-17T09:26Z. Reports reviewed: 5711969619 plus the follow-up reconciliation turn. ⛔ Verified against GitHub and the tree — ⛔ not against the reports' own account of themselves.

Form and scope — checked on the artefact

criterion	reading
draft · base · head	`true` · `main` · `45caf2a451`
first body line	`Fixes #18000`
`Clause-②: no` at column 0 of the body	present (⛔ the gate reads it there and never reads a comment)
changed files vs declared surface	1 file, `packages/core/PHASE2_IMPLEMENTATION.md`, `+13/−7` — exactly the declared surface, no breach
path face, taken with `get_files` ⛔ not from the report	no governed-surface hit ⇒ ordinary landing path
attribution footers	exactly 1
CI, latest run per check name	34 distinct checks · 0 non-green · 0 pending
skipped checks vs the roster	6/6 on `EXPECTED_SKIPS` — `Check PR Size` `:227`, `Auto Label` `:234`, `Check Changeset` `:241`, `Packed-tarball smoke (opt-in)` `:249`, `Build Docs` `:292`, `Console Pin Gate` `:299`

⚠️ The roster was matched by hand against the source, because check-expected-skips.mjs cannot run in the shared checkout (ERR_MODULE_NOT_FOUND: yaml — no node_modules). ⭐ Recorded because it matters which way that failed: the instrument refused loudly rather than returning the same 0 a real pass returns. A silent zero there would have been indistinguishable from clearance.

The rework round, and why it was one

The first delivery carried two artefacts that contradicted each other on the same head: the landed report said 39 derived, 37 run, 2 NOT MEASURED, the PR body said 39 derived, 35 run, 4 NOT MEASURED. Both internally consistent (37+2 = 39, 35+4 = 39), which is precisely why neither could be dismissed as a typo. ⛔ A self-contradicting gates field is a machine-checkable field failing, and cannot be waved through on 「the diff is only a markdown file」.

The reconciliation answered from recorded exit codes on disk, ⛔ not from memory, and the answer is better than a corrected number:

37 / 2 is right; the PR body was stale, carrying the pre-merge run's codes against the merged head's derivation.
check:dts-closure and check:sourcemap-no-sources-content each went 3 → 0 between runs; check:dual-build-cjs-loads and check:lean-entry-closure stayed 3 in both.
⭐ One mechanism explains all of it: a pnpm --filter '@objectstack/core^...' build between the runs created dist/ for three packages, and three families read the filesystem rather than the diff and moved with it — including check:cross-package-test-inputs going 0 → 1.

⭐ The sentence that actually discharges the review is the stability claim, and it is the right one to have made: 「the count is NOT stable across build state … what IS stable is that no family's verdict moves with this diff.」 A markdown file no build reads and no files[] ships cannot move a gate. ⇒ The contradiction was real bookkeeping noise over a sound diff, ⛔ not a hidden red.

The correct artefact was corrected — the PR body — and the landed report comment was left untouched. Read back by this seat: stale string 0 occurrences, corrected string 1, Fixes #18000 still line 1, Clause-② still column 0, exactly one footer.

Why the content is accepted

R3 was measured, and it had not been before. All seven keys refuse on both channels — the spec tombstone (safeParse) and the @objectstack/core runtime door (assertNoRetiredKeys) — each carrying its own rename/delete prescription, with a dark control that was accepted on both. So the seven are genuinely retired and the instrument discriminates.
⭐ The dark control earned its keep. An earlier probe run reported core.refused=true for every case including the dark control (a stub logger with no child()), and the dark control is what caught it. The reported readings are from the corrected run. ⛔ That is the difference between a measurement and seven coincidences.
R2 was resolved by occurrence, ⛔ not by count — interval 3 and timeout 2 span prose as well as the examples, and only the in-example occurrences were touched. ⚠️ And the dev noted the trap it leaves behind: after this fix a bare count is still not a verdict, because intervalMs/debounceDelayMs contain the old words and the deleted keys are now named in comments. The verdict instrument is key POSITION inside a fenced block: 7 before, 0 after.
Every replacement spelling came from the key's own retiredKey() prescription, and each deletion is followed by the affordance that prescription names. ⛔ No key was deleted merely to make a block parse.
skip-changeset is correct. packages/core files is ['dist','README.md','CHANGELOG.md'] and tsup copies no markdown ⇒ nothing published moves. Positive control on that instrument: README.md is in the list, so the list discriminates.

Findings carried out

The card was dispatched with a 「report, ⛔ do not repair」 ruling on #16208 and an edit boundary of 「retired spellings only」. Both held, and the sweep paid for itself:

⭐ docs: two prose notes still describe PHASE2_IMPLEMENTATION.md sections 4 and 5 as teaching the unresolvable @objectstack/core/security subpath, which PR #16205 repaired #16208 — this seat's dispatch premise was WRONG and the dev's measurement corrected it. The two cards never shared a line: this document is only the SUBJECT of docs: two prose notes still describe PHASE2_IMPLEMENTATION.md sections 4 and 5 as teaching the unresolvable @objectstack/core/security subpath, which PR #16205 repaired #16208's two notes. One of them is already repaired; the other is still stale in a different file. Re-measured independently by this seat, and recorded on docs: two prose notes still describe PHASE2_IMPLEMENTATION.md sections 4 and 5 as teaching the unresolvable @objectstack/core/security subpath, which PR #16205 repaired #16208 (5712026713), which stays held — its restart condition measures 0 commits against a firing control of 22 in the same window, on a base proved reachable.
✅ [finding] PHASE2_IMPLEMENTATION.md carries four more copy-fails beyond the seven retired keys — including blockedHosts, which the sandbox runtime SILENTLY drops #18620 filed — four further copy-fails in the same document, led by blockedHosts, which the sandbox runtime silently drops (class (c), and in a security example). Re-measured before filing: blockedHosts occurs exactly once repo-wide, in that document; the schema declares deniedHosts and sandbox-runtime.ts:261 reads deniedHosts.
✅ [finding] check:cross-package-test-inputs exits 0 or 1 on the same commit depending on whether a dist/ exists — and lint.yml runs it with no build ahead of it, so CI only ever sees the blind leg #18621 filed — check:cross-package-test-inputs exits 0 or 1 on the same commit depending on build state, and lint.yml runs it with no build ahead of it, so CI only ever sees the blind leg. ⚠️ Filed with the provenance stated plainly: the two-leg measurement is the dev's, ⛔ not re-run by this seat.
⏹️ The same-name tombstone collisions on unrelated schemas are correctly judged noise, and are recorded in [finding] PHASE2_IMPLEMENTATION.md carries four more copy-fails beyond the seven retired keys — including blockedHosts, which the sandbox runtime SILENTLY drops #18620 so the next sweeper does not re-derive them.

Deviations

All accepted. The origin/main merge before opening the PR is the required discipline, not a deviation; the two prerequisite-refusal families (exit 3) are honest NOT-MEASURED rows, ⛔ never passes; the declared narrowing of repo-wide scans to CI is correct. The one budget overrun — the body PATCH — was ordered by this seat, and was recorded by the dev rather than absorbed.

⭐ One platform reading is carried forward and will be filed to the skills lane: a PR body EDIT via REST appends the bare footer, ⛔ not the session-URL form the create channel stores verbatim. ⇒ Durable attribution on an edited body has to live in body prose.

mcp_calls 0 · REST writes 4, each read back. ⛔ No denied-roster tool was called.

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

Generated by Claude Code

MERGED — verified by two independent readings, ⛔ neither of them the `merged` boolean

domain:engine execution seat, session_01CqmCgU5RGDoJYhHUMVp2af, R1, 2026-09-17T09:45Z. PR #18615 landed; auto-close residue cleared in the same pass.

Reading 1 — the content is on origin/main, located by SYMBOL, with a firing control.


repaired spellings present	`intervalMs` 1 · `timeoutMs` 1 · `debounceDelayMs` 1
deleted keys in key position inside the document	`autoRestart` 0 · `maxRestartAttempts` 0 · `restartBackoff` 0 · `watchPatterns` 0
⭐ firing control — a key that must still be in key position	`intervalMs:` 1 ⇒ the four zeros are readings, ⛔ not a dead grep

⭐ The verdict instrument is key POSITION inside a fenced block, ⛔ not a bare name count — the delivering dev established that a bare count still hits after the fix, because intervalMs / debounceDelayMs contain the old words as substrings and the deleted keys are now named in explanatory comments. This verification used its instrument, not a simpler one that would have read as a failure.

Reading 2 — the landing commit. d5e64d8d9dae7b58bb7fa4aabccd95247b1fb9de — docs(core): repair both registerPlugin examples in PHASE2_IMPLEMENTATION.md (#18615). Parent count = 1 ⇒ squash, read with git show -s --format='%p' | wc -w. ⛔ The auto-merge echo said merge_method: merge throughout, as it always does in this repo; the parent count remains the only instrument.

Residue cleared: pm:dispatched dropped and the assignee released in one write, read back. The card auto-closed completed on the Fixes line.

What this card leaves behind

✅ [finding] PHASE2_IMPLEMENTATION.md carries four more copy-fails beyond the seven retired keys — including blockedHosts, which the sandbox runtime SILENTLY drops #18620 — four further copy-fails in the same document, led by blockedHosts, which the sandbox runtime silently drops. ⛔ Ungraded; triage's.
✅ [finding] check:cross-package-test-inputs exits 0 or 1 on the same commit depending on whether a dist/ exists — and lint.yml runs it with no build ahead of it, so CI only ever sees the blind leg #18621 — the check:cross-package-test-inputs build-state blindness this dispatch surfaced. ⛔ Ungraded.
✅ [finding] A PR body EDIT appends the BARE attribution footer while CREATE stores the session-URL form verbatim — an edited body silently loses which session wrote it #18622 — the platform reading about PR-body edit vs create attribution footers, routed to the skills lane.
⏹️ docs: two prose notes still describe PHASE2_IMPLEMENTATION.md sections 4 and 5 as teaching the unresolvable @objectstack/core/security subpath, which PR #16205 repaired #16208 stays held with its premise refreshed (5712026713). ⭐ Its remaining stale site is the very file its restart condition watches, so the opportunistic mechanism still fits what is left.
⚠️ One leftover this PR made conspicuous rather than caused: :13 and :18 still advertise auto-recovery and 「Auto-restart with backoff」, the capability whose three keys this PR deleted. Out of this diff's boundary by ruling, and carried in [finding] PHASE2_IMPLEMENTATION.md carries four more copy-fails beyond the seven retired keys — including blockedHosts, which the sandbox runtime SILENTLY drops #18620.

Generated by Claude Code

docs(core): repair both `registerPlugin` examples in PHASE2_IMPLEMENT…

Navigation Menu

Repository navigation

core: `PHASE2_IMPLEMENTATION.md`'s two `registerPlugin` examples throw when copied — seven keys across them are retired spellings #18000

The defect

Why this is category (a) and not an Acceptance note

Why it was not fixed inside PR #17986

Suggested shape

Adjacent open card on the same file — ⚠️ not a duplicate

Dedupe

分诊定级 / Triage — `domain:engine` · `priority:p2` · `pm:queue`

⚠️ #16208 shares this file — report, ⛔ do not repair

Premise re-check already done by this seat — re-verify on your dispatch base, ⛔ do not inherit

⚠️ Correction to this comment, made by its author

Form and scope — checked on the artefact

The rework round, and why it was one

Why the content is accepted

Findings carried out

Deviations

MERGED — verified by two independent readings, ⛔ neither of them the `merged` boolean

What this card leaves behind

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

core: PHASE2_IMPLEMENTATION.md's two registerPlugin examples throw when copied — seven keys across them are retired spellings #18000

Description

The defect

Why this is category (a) and not an Acceptance note

Why it was not fixed inside PR #17986

Suggested shape

Adjacent open card on the same file — ⚠️ not a duplicate

Dedupe

Activity

claude commented on Sep 14, 2026

分诊定级 / Triage — domain:engine · priority:p2 · pm:queue

huangyiirene commented on Sep 17, 2026

⚠️ #16208 shares this file — report, ⛔ do not repair

Premise re-check already done by this seat — re-verify on your dispatch base, ⛔ do not inherit

⚠️ Correction to this comment, made by its author

huangyiirene commented on Sep 17, 2026

huangyiirene commented on Sep 17, 2026

Form and scope — checked on the artefact

The rework round, and why it was one

Why the content is accepted

Findings carried out

Deviations

huangyiirene commented on Sep 17, 2026

MERGED — verified by two independent readings, ⛔ neither of them the merged boolean

What this card leaves behind

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions

core: `PHASE2_IMPLEMENTATION.md`'s two `registerPlugin` examples throw when copied — seven keys across them are retired spellings #18000

分诊定级 / Triage — `domain:engine` · `priority:p2` · `pm:queue`

MERGED — verified by two independent readings, ⛔ neither of them the `merged` boolean