docs(adr): declared journeys as the priority anchor — the maintainer's ≤10 end-to-end paths, each with its runner, become the first triage question (maintainer 「立」)

Filed and dispatched by the director seat (summon #24, session_01Wj1HUjzyeiBQ8atRf1ZhaL) under the maintainer's ruling on the direct channel, 2026-09-16 ~14:5xZ: after 「我觉得一个创业项目核心应该放在怎么解决实际业务需求上,你再帮我综合评估我们最近的开发还有哪些跑偏了」, 「我们是一个开发平台,也是创业项目,用的人少,你也不会知道『谁今天撞上』」 and 「路的清单是否建议重新讨论,专门立文档」 — reply to the ADR proposal, verbatim: 「立」. ⛔ No domain:* set by the filer; docs/adr/** is the governed face (domain:skills); the director dispatches directly.

Dedup terms: declared journeys · journey-anchored priority · 路的清单 · 黄金旅程 ADR · priority anchor

Why (director's readings, 2026-09-16)

Four repos, 1,166 merges since 09-02 (shallow floor): fix 36% · process/gates 30% · docs 16% · feature 8% · tests 6%. Of objectstack's 86 feat merges in 30 days, most are contract declarations, tombstones, refusals and gates; user-facing capability landed in the dozens, not hundreds. Meanwhile cloud's six P0s (three credential leaks, production signup broken 22 d, control plane unlaunchable 26 d, a 15 h outage) sit in waiting states, and hotcrm's one customer requirement line (REQ-0006) was blocked by a code-size ratchet. Queue: spec 88 · devx 57 vs engine 9 · services 1 · cli 3. The protocol's first triage question is 「违背了哪条契约」; with few users no pull reading exists, so the maintainer's own declared end-to-end paths are the anchor instead.

Deliverable — ONE ADR, draft, the maintainer edits by deletion

docs/adr/0136-declared-journeys-as-priority-anchor.md (take the next free number on the tree at claim time), in the repo's ADR shape (Status: Proposed, context / decision / consequences), short:

The journey table — ≤10 rows, drafted from existing sources, ⛔ nothing invented

One row per declared end-to-end path, five columns: who walks it (tenant admin / app author / AI app builder / platform operator) · entry → what the user sees when it is done · runner (the mechanical proof: a docs/qa/platform-checklist/ item id, a step of cloud scripts/dev-local/MAGIC-FLOW-TEST-RUNBOOK.md §7 「Test checkpoints」, a dogfood script, or none yet) · last reading (🟢/🔴 + date + the card if red) · source (ADR-0111 row, REQ number, epic card).

Sources to draft from, all on origin/main: cloud docs/adr/0111-hotcrm-hosted-saas-mvp-services-model.md acceptance rows 1.1–5.x; cloud scripts/dev-local/MAGIC-FLOW-TEST-RUNBOOK.md §7 (the golden-journey checkpoints); cloud#1521 (control plane launchable) and cloud#1653 (new-user workspace); hotcrm docs/requirements/0001–0006 (the REQ line, one row for 「a customer requirement lands end-to-end in HotCRM」); objectstack docs/qa/platform-checklist/ (areas + coverage.json) for the platform-side paths (author an app, install a package, AI builds an app from a prompt). Merge duplicates; if the honest count exceeds 10, keep the 10 with a runner and list the rest under 「未入选」 with one line each so the maintainer can swap.

The decision section — three rulings, stated as rules the PM protocol will cite

Journey-anchored priority. Every triaged card names the journey it sits on (Journey: line). A journey broken or blocked ⇒ P0/P1. A journey that runs but produces a wrong result on a path it exercises (incl. an AI-authored app silently swallowed by the platform) ⇒ P2. On no declared journey ⇒ p3 or not_planned by default — a declared-but-unenforced key is retired in batches per release, ⛔ not one card per key. Security and data integrity never wait for a journey: always the top band.
Cross-lane priority. While any product repo (cloud / hotcrm / objectui user-facing) has an unclaimed P0/P1 on a journey, no lane dispatches a p2/p3 contract-hygiene or tooling card. One sentence; the PM protocol's 取卡全序 cites it.
The queue is fed by running the journeys. The runners are executed on a schedule (the existing checklist-test / dogfood-verification skills); a red journey files a card with its Journey: line. Contract sweeps and findings become background, graded against the journey list.

The ADR states in one line that it supersedes docs/adr/PRIORITIZATION.md (a 2026-06 ADR-status review, self-marked STALE) as the priority anchor; that file gains a one-line pointer, ⛔ not rewritten.

Consequences section

The list is the maintainer's appetite: the maintainer edits it by deletion and swap; a seat proposes a new row only with a runner attached. Ten is the cap.
The SKILL.md triage lines that implement rulings 1–3 are a separate skills-lane card filed after this ADR is accepted (⛔ not in this PR; the ADR is the law, the skill edit is the enforcement).

Acceptance

≤10 rows, every row with a real source citation and either a runner path/id that exists on the tree or none yet; the dev's report lists the four source documents' row counts and how the table was reduced.
ADR file in the repo's shape; docs/adr/PRIORITIZATION.md +1 line pointer; nothing else touched.
Governed (docs/adr/**): draft PR, ## 维护者速读(草稿), four-piece by the director, the maintainer's approval lands it. The maintainer is expected to DELETE rows in review — the PR body says so.

Refs objectstack#18472 (gate census) · #18473 (asymmetry flip) · cloud#1521 · cloud#1653 · hotcrm docs/requirements/README.md

Generated by Claude Code

Claim: PM loop round 1 — maintainer direct dispatch to the director seat
Session: session_01Wj1HUjzyeiBQ8atRf1ZhaL
Branch: claude/issue-18477-declared-journeys-adr
Worktree: objectstack-issue-18477
Domain: domain:skills
File surface: docs/adr/0136-declared-journeys-as-priority-anchor.md (new; number re-taken at claim from the tree) · docs/adr/PRIORITIZATION.md (+1 pointer line) (stop on breach; explain in the report)
Container & model: M (drafting from four existing sources; no code), mode:subagent, model: claude-fable-5-1 — dispatch-gates.mjs --tier: no path-derived mandate for docs/adr/**; the director dispatches at the session's own tier because the ADR is the maintainer's priority law and will be edited by deletion; no downgrade exit taken
Clause-②: no
Thread-read: none
Serial constraints cleared: no open PR touches docs/adr/0136* or PRIORITIZATION.md (git ls-remote --heads origin 'claude/issue-1847*' lists only #18470 / #18472 / #18473 / #18468 branches, none on docs/adr/**); the three director dispatches in flight touch .claude/**, scripts/pm/*, docs/audits/** — disjoint; os-verify-lock.sh --status read no holder, no tickets at 14:2xZ

Authority for a director claim: the maintainer's ruling on this session's channel, verbatim 「立」 (2026-09-16 ~14:5xZ) on the proposal to file the declared-journeys ADR, and the standing instruction that the director dispatches these cards directly.

Generated by Claude Code

os-dev-report

{
  "issue": 18477,
  "status": "done",
  "branch": "claude/issue-18477-declared-journeys-adr",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/18480",
  "session": "session_01Wj1HUjzyeiBQ8atRf1ZhaL",
  "premise_still_valid": true,
  "summary": "ADR-0136 written at docs/adr/0136-declared-journeys-as-priority-anchor.md (0136 was the first free number on origin/main 85c6d76e at claim time, as the card predicted) and ONE pointer line added at the top of docs/adr/PRIORITIZATION.md; nothing else touched, SKILL.md untouched. Journey table: 9 rows (cap 10), 7 with a runner, 2 `none yet` (J2 control-plane launch speed: the six cloud#1521 criteria are prod-only manual measurements; J6 customer requirement landing end-to-end in HotCRM: pnpm verify and e2e specs do not prove a REQ landed), plus 5 未入选 lines. Every runner cited was checked to exist on its tree (git cat-file / ls-tree on each repo's origin/main). Sources and counts before reduction: cloud ADR-0111 17 acceptance rows (R1–R6; the card said 1.1–5.x = 15) → J5 merges R1–R3 (11 rows), R5.2 → J1, R4/R6.1/R6.2 → 未入选, R5.1 → security band; cloud MAGIC-FLOW-TEST-RUNBOOK.md §7 8 checkpoints → J3 (4) + J4 (4); cloud#1521 (6 criteria) → J2, cloud#1653 → J1; hotcrm docs/requirements 6 records (0001 illustrative, 0002–0006 one customer's Track A, all Triaged) → J6; objectstack platform-checklist 264 items / 15 areas / 19 P0 → J7, J8, J9 cite 7 items, 未入选 3 more. Total 302 candidate rows → 9 journeys. Decision section carries the three rulings as rules (journey-anchored bands with security/data-integrity always top; cross-lane priority; queue fed by running the journeys) and the one-line supersession of PRIORITIZATION.md. Draft PR #18480 body carries `## 维护者速读(草稿)` in Chinese (five parts, 席位意见 empty), says the maintainer is expected to delete and swap rows, lists 9 rows / 7 runners / 2 none-yet, and names the four source documents with their counts. Assignee was already set by the dispatch (hotlong); newest Claim: names this branch and session — verified, no second claim posted. Governed surface: PR left DRAFT, not flipped, nothing armed.",
  "journeys": "rows 9; with runner 7 (J1, J3, J4, J5, J7, J8, J9); none yet 2 (J2, J6); readings: 🔴 J1 cloud#1653 (2026-09-16), 🔴 J2 cloud#1521 (2026-08-21 readings, awaiting prod re-measure 2026-09-16), 🟢 J3 golden-journey nightly run 17 2026-09-15, 🟢 J4 2026-08-01 (time-relative leg, runbook), 🟢 J5 2026-08-16 rig hard checks (3 pinned findings), 🔴 J6 0 of 6 REQ past Triaged (hotcrm@087b7c5), 🟢 J7 scaffold-e2e run 3082 2026-09-16 + unknown for first-run-loop, unknown J8, 🟢 J9 Dogfood Regression Gate on main@85c6d76e 2026-09-16 (API half) + unknown browser half",
  "files_changed": "docs/adr/0136-declared-journeys-as-priority-anchor.md (new, +88 lines); docs/adr/PRIORITIZATION.md (+1 line, line 2 pointer); commit dc2b2ff3 on origin",
  "tests": "No package touched, so no ① dependency-closure build and no ② package test/typecheck are owed; repo-wide scans are CI's. Gates: derived with `node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack docs/adr/0136-declared-journeys-as-priority-anchor.md docs/adr/PRIORITIZATION.md` = 18 commands; all run with exit captured to a file before any pipe (tree dc2b2ff3): check-adr-links :: 0 (`683 relative link destination(s) under docs/adr/ resolve`) · check-adr-links --self-test :: 0 · check-adr-symbol-anchors :: 0 (`2118 anchors across 140 records resolve`) · --self-test :: 0 · check-ci-filter-parity :: 0 · check-closing-keyword-parity :: 0 · --self-test :: 0 · check-comment-mask-corpus :: 0 · pnpm check:adr-anchors :: 0 (only the pre-existing allowlisted ADR-0057 note) · check:cross-package-test-inputs :: 0 · check:doc-authoring :: 0 (`402 files clean`) · check:driver-memory-census :: 0 · check:nul-bytes :: 0 (`scanned 8735 text file(s) ... no raw ASCII control bytes`) · check:pm-governed-merges :: 0 · check:pm-prior-rulings :: 0 · check:refd-timer-probe :: 0 · check:watch-hint-literal :: 0 · pnpm --filter @objectstack/lint run check:doc-formula-expressions :: first exit 3 = PREREQUISITE NOT MET (formula/lint unbuilt in the fresh worktree; by the gate's own text nothing measured, not a finding) → built both under `os-verify-lock.sh` (VERDICT command-exit 0, held 1s, waited 0s) → re-run :: 0. Re-derived from the merge base after the commit and reconciled: `dispatch-gates --ran: 18 derived, 18 run, 0 NOT-MEASURED (a DERIVED zero — all 18 recorded an exit code and none of them is 3)`. The tool also names 53 artifact-roster, 11 wide-population and 14 pending-changeset families as outside the derived total — CI's. PR CI: in_progress at report time (not waited on). No ablation applies (prose-only change).",
  "mcp_calls": "0 — no MCP GitHub tool was called (reads: curl GET via REST proxy on objectstack#18477 + comments, cloud#1521/#1653 + comments, cloud golden-journey.yml runs, hotcrm e2e.yml/ci.yml runs, objectstack scaffold-e2e.yml runs, objectstack commits/85c6d76e/check-runs, pulls/18480 read-back; sibling sources read with git show origin/main in /home/user/cloud and /home/user/hotcrm, read-only)",
  "api_writes": "3 REST writes — POST /repos/objectstack-ai/objectstack/pulls ×1 (201 → #18480, draft; body read back byte-identical, one session-URL footer); POST /repos/objectstack-ai/objectstack/issues/18480/labels ×1 via scripts/pm/label-write.mjs (200; read-back `size/s`, `skip-changeset` — the size-labeler's `size/s` was already present and the additive write kept it); POST /repos/objectstack-ai/objectstack/issues/18477/comments ×1 (this report). PATCH /pulls ×0. Plus git push ×2 (empty-branch probe exit 0, then dc2b2ff3).",
  "deviations": "(1) J1's runner cites cloud scripts/dev-local/LOCAL-E2E-CHECKLIST.md A2–A3 · B2 and J4 cites its B8 as a secondary runner; the 未入选 publisher-flywheel line is sourced from the same file (D1–D6). It is an on-tree checklist beside the named runbook but not one of the four named source documents; the alternative was `none yet`. Declared in the PR body for the maintainer to strike. (2) Readings for J3, J7, J9 come from read-only REST reads of Actions runs / check-runs on 2026-09-16, not from the named documents (which carry no reading for those rows). (3) check:doc-formula-expressions exit 3 → prerequisite build under the lock → 0, as above. (4) The dispatch's write list did not enumerate the labels POST; the agent file's budget does and says to label on PR open, so it was done and is counted above. (5) The ADR body is in Chinese with English identifiers, as the dispatch allowed; row text kept terse. No line-length lint applies to docs/adr (max line in the file 595 bytes, table rows; existing ADRs reach 642).",
  "open_questions": [],
  "out_of_scope_findings": [
    "noted, not filed: the card names cloud ADR-0111 acceptance rows 1.1–5.x, but the file on cloud origin/main (7be90d0) also carries R6 (6.1–6.2), so the honest count is 17 rows, not 15. 承接者: the maintainer editing the table in PR #18480."
  ]
}

Generated by Claude Code

ACCEPT — director seat, session_01Wj1HUjzyeiBQ8atRf1ZhaL, served tier claude-fable-5-1 · 2026-09-16T15:10Z

Verified against GitHub (PR #18480, head dc2b2ff3, draft, 2 files +89/−0), ⛔ not against the report:

Scope held: docs/adr/0136-declared-journeys-as-priority-anchor.md (new, 88 lines) + one pointer line at the top of docs/adr/PRIORITIZATION.md; nothing else; SKILL.md untouched (the enforcement is the follow-up skills-lane card, as the card required).
Table: 9 rows (cap 10) — J1 new-user signup→console (🔴 cloud#1653) · J2 control-plane launch speed (🔴 cloud#1521, none yet) · J3 AI builds an app from a prompt (🟢 nightly golden-journey) · J4 AI adds an automation (🟢 2026-08-01) · J5 HotCRM hosted tenant isolation (🟢 rig) · J6 a customer requirement lands end-to-end (🔴 0/6 REQ past Triaged, none yet) · J7 zero-code author scaffold→publish (🟢 scaffold-e2e) · J8 marketplace install (unknown) · J9 console CRUD + row visibility (🟢 Dogfood gate). Every cited runner was checked to exist on its tree (the report names the git cat-file / ls-tree reads); 5 未入选 lines carry what was left out and why. Sources reduced from 302 candidate rows and the counts are stated (ADR-0111 has 17 rows, not the 15 the card said — the dev's correction is right).
Decision: D1 table; D2 the three rulings as rules (journey bands with security/data-integrity always top; cross-lane priority; queue fed by running the journeys); D3 supersession. Consequences name the cost honestly (D2.2 stalls hygiene cards while a product P0/P1 is open — intended).
Gates: 18 derived / 18 run / 0 unrun (--ran, DERIVED zero), all 0 after one prerequisite build (check:adr-links 683 destinations resolve, check:adr-anchors, check:doc-authoring 402 clean); mcp_calls 0; api_writes 3 REST + 2 pushes, read back; skip-changeset read back. CI on dc2b2ff3 at review: 18 success · 11 skipped · 4 in progress · 0 failed.
Deviations answered: J1/J4 cite LOCAL-E2E-CHECKLIST.md (on-tree, beside the named runbook) as a runner rather than none yet — accepted, and flagged in the 速读 for the maintainer to strike if unwanted; readings for J3/J7/J9 come from Actions/check-run reads dated today — accepted, the ADR says readings are snapshots.

Landing (governed docs/adr/**, ruling C): four-piece hung on PR #18480. ⚠️ This PR is the one the maintainer is expected to EDIT (delete / swap rows) before approving; the seat lands whatever head carries the approval. Card stays pm:dispatched until MERGED; the SKILL.md enforcement card is filed on acceptance.

Implemented-by: claude/issue-18477-declared-journeys-adr (mode:subagent)
Reviewed-by: session_01Wj1HUjzyeiBQ8atRf1ZhaL

Generated by Claude Code

Ruling on the draft — REWORK: the journey table is the DEVELOPER PLATFORM's, not the cloud service's · maintainer (os-zhuang), review 5224841038 on PR #18480, verbatim: 「这是云服务的旅程,不是本应用开发平台的旅程。本平台的目的是不能让第三方开发者能进入我们的元数据协议快速地开发迭代企业管理应用。」 2026-09-16T15:36Z

Director seat, summon #24, session_01Wj1HUjzyeiBQ8atRf1ZhaL. The maintainer approved PR #18480 at 2026-09-16T15:32:50Z and then, on reading the table, requested changes at 2026-09-16T15:34Z with the sentence above. The seat reads 「不能让」 as 「能让」 from context (the sentence states the platform's purpose); the direction is unambiguous either way: the anchor is the third-party developer entering the metadata protocol and building / iterating an enterprise management application quickly. The earlier approval is superseded by the changes-requested review; the label was already stripped, and the four-piece is re-hung after the rework.

What the rework changes (D1 only; D2 rulings and D3 supersession stand)

The table is re-centred on the developer journey, end to end: a third-party developer discovers the protocol (docs site, published skills/objectstack-*, create-objectstack) → scaffolds an app → authors objects / fields / views / flows / permissions as metadata → runs it locally (os dev) and sees it in the console → verifies (pnpm verify, the AI-authoring traps refused loudly) → packages and publishes → installs it into an environment → iterates a change against a customer requirement and sees it live (hotcrm's REQ line is the reference third-party app doing exactly this). AI-assisted authoring (the AI writes the metadata) is a developer journey, not a cloud one.
The app's end users appear only as the proof that the built app works (records CRUD, permissions visible in the UI), not as the anchor.
Cloud-service rows (account signup, control-plane launch speed, hosted HotCRM tenancy / isolation, billing posture) leave the table: they move to 「未入选 — cloud 服务自己的清单」 with one line each, and the ADR says the cloud repo may declare its own list under the same shape if the maintainer wants one. objectui (the rendering half of the developer's app) is part of the developer journey, not a separate anchor.
Sources stay on-tree only: objectstack docs/qa/platform-checklist/ (studio-authoring, cli.scaffold, platform-core.marketplace-install, records-forms, permissions areas), packages/create-objectstack, packages/cli docs, skills/objectstack-* READMEs, content/docs/ getting-started, hotcrm docs/requirements/; cloud's runbook only where it proves a developer step (AI builds an app from a prompt).
Cap stays ≤10; every row keeps a runner or none yet.

Execution

The same dev continues on the same branch (patch round, not a re-dispatch); the PR body's 维护者速读(草稿) is rewritten to say the table is now the developer platform's journeys and to name what moved to 未入选.
The director re-reviews, re-hangs the four-piece and asks the maintainer to read the table again. Card stays pm:dispatched.

Generated by Claude Code

os-dev-report

{
  "issue": 18477,
  "status": "done",
  "branch": "claude/issue-18477-declared-journeys-adr",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/18480",
  "session": "session_01Wj1HUjzyeiBQ8atRf1ZhaL",
  "premise_still_valid": true,
  "summary": "Rework round per the director's ruling comment (2026-09-16T15:36Z) and the maintainer's changes-requested review 5224841038 (verbatim 「这是云服务的旅程，不是本应用开发平台的旅程。本平台的目的是不能让第三方开发者能进入我们的元数据协议快速地开发迭代企业管理应用。」). D1 rewritten: the table is now the DEVELOPER PLATFORM's journeys — 10 rows (cap), 9 with a runner, 1 `none yet` (J9). J1 discover + scaffold (getting-started → `npm create objectstack` → 10 skills → health) · J2 author metadata by hand · J3 AI-assisted authoring (Claude Code + skills, or a prompt to the build agent; cloud runbook §7 / golden-journey kept only as this row's runner) · J4 Studio zero-code loop · J5 `os dev` and see it in the console (objectui = rendering half, inside the row) · J6 verify (`os validate` / `os lint` / `objectstack verify`, traps refused loudly) · J7 package + publish · J8 install into an environment · J9 iterate against a customer requirement (hotcrm `docs/requirements/` + `.github/tasks/new-feature.md`) · J10 the app's end users as the closing proof (CRUD, RLS, FLS), one row only. Moved out to 「未入选 — cloud 服务自己的清单」 (6 lines): new-user signup → org → env (cloud#1653); control-plane launch speed (cloud#1521); hosted HotCRM login / per-org space / isolation (cloud ADR-0111 R1–R3); operator enablement + blast radius (R4, R6.1); billing posture (ADR-0111 D1); cloud-side review + counter-sign (LOCAL-E2E D5–D6); plus one sentence that the cloud repo may declare its own list in the same shape. 「未入选 — 平台侧」 (2 lines): approvals-inbox entry; custom-page items folded into J2/J4. D2 (three rulings, now J1–J10), D3 and the Consequences stand; the source/merge accounting follows the new table. Sources (all on-tree, nothing invented): objectstack platform-checklist 264 items / 15 areas / 19 P0 → 22 items in the table + 3 in 未入选; `content/docs/getting-started/` 8 pages → J1, J3; `skills/objectstack-*` 10 → J1–J3; `packages/create-objectstack/README.md` (1 template) → J1; `packages/cli/README.md` command tables → J5–J7; `packages/verify/README.md` → J6, J10; `packages/mcp/README.md` → J3; hotcrm `docs/requirements/` 6 records + `.github/tasks/new-feature.md` + 4 workflows → J6, J7, J9; cloud runbook §7 8 checkpoints → J3 runner only; first-draft cloud sources (ADR-0111 17 rows, cloud#1521, cloud#1653, LOCAL-E2E A/D) → moved out. `origin/main` moved (85c6d76e → 9c1897e5); merged with `bash scripts/pm/os-regen-merge.sh` (merge commit 082b9e37, nothing to regenerate), then the rework commit 365d1bc0; three-dot diff vs origin/main is still exactly the two files (+114). PR #18480 stays DRAFT, nothing flipped or armed; body rewritten via one PATCH and read back. Every newly cited runner path was checked to exist on its tree (`git cat-file -e` on each repo's origin/main).",
  "journeys": "rows 10; with runner 9 (J1–J8, J10); none yet 1 (J9); moved out 6 cloud lines + 2 platform lines. Readings (2026-09-16 snapshots): 🟢 J1 scaffold-e2e run 3082 + check-links run 9879; 🟢 J2 Dogfood Regression Gate on main@85c6d76e (coverage ratchet unknown); 🟢 J3 cloud golden-journey nightly run 17 (2026-09-15); J4 first-run-loop unknown / draft-publish leg 🟢; J5 unknown (manual items); 🟢 J6 hotcrm ci run 3054 on 087b7c5 + Dogfood; 🟢 J7 hotcrm publish-staging run 11 + publish-production run 5 (2026-09-14); J8 unknown; 🔴 J9 0 of 6 REQ past Triaged (app itself 🟢 ci 3054 / e2e 1272); 🟢 J10 Dogfood gate (API half), browser half unknown",
  "files_changed": "docs/adr/0136-declared-journeys-as-priority-anchor.md (rewritten, now 113 lines; +113 vs origin/main); docs/adr/PRIORITIZATION.md (+1 line, unchanged from round 1). Commits on the branch since round 1: 082b9e37 (merge of origin/main 9c1897e5), 365d1bc0 (rework). Remote head 365d1bc0.",
  "tests": "No package touched, so no ① dependency-closure build and no ② package test/typecheck are owed; repo-wide scans are CI's. Gates: derived with `node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack docs/adr/0136-declared-journeys-as-priority-anchor.md docs/adr/PRIORITIZATION.md` on tree 365d1bc0 = 18 commands (identical set to round 1); all run with exit captured to a file before any pipe: check-adr-links :: 0 · --self-test :: 0 · check-adr-symbol-anchors :: 0 · --self-test :: 0 · check-ci-filter-parity :: 0 · check-closing-keyword-parity :: 0 · --self-test :: 0 · check-comment-mask-corpus :: 0 · pnpm --filter @objectstack/lint run check:doc-formula-expressions :: 0 (formula + lint built first under `os-verify-lock.sh`, VERDICT command-exit 0, waited 0s) · pnpm check:adr-anchors :: 0 · pnpm check:cross-package-test-inputs :: 1 (see below) · check:doc-authoring :: 0 · check:driver-memory-census :: 0 · check:nul-bytes :: 0 · check:pm-governed-merges :: 0 · check:pm-prior-rulings :: 0 · check:refd-timer-probe :: 0 · check:watch-hint-literal :: 0. Reconciled from the merge base 9c1897e5: `dispatch-gates --ran: 18 derived, 18 run, 0 NOT-MEASURED (a DERIVED zero — all 18 recorded an exit code and none of them is 3)`. The one red: `check:cross-package-test-inputs` names `packages/cli/test/init-created-files-summary.e2e.test.ts` walking into `packages/spec/dist/` (no declared glob reaches it). That directory exists in the worktree only because the lock-wrapped formula/lint build built `@objectstack/spec` as a dependency; in round 1 the same gate ran BEFORE that build on the same code and exited 0; stable on re-run (exit 1); `origin/main` tip 9c1897e5 shows `Lint & Repo Gates` success at 2026-09-16T15:25Z on the same gate. This diff touches no test, no turbo.json, no package — the red is a build-state property of an untouched test, filed below as an out-of-scope finding. PR CI: in_progress at report time (not waited on). No ablation applies (prose-only change).",
  "mcp_calls": "0 — no MCP GitHub tool was called (reads this round: curl GET via REST proxy on issues/18477/comments, pulls/18480 + reviews + review comments, cloud golden-journey runs, hotcrm e2e/ci/publish-staging/publish-production runs, objectstack scaffold-e2e / check-links / docs-drift-check runs, commits/9c1897e5/check-runs, pulls/18480 read-back, issues/comments read-back; sibling sources via git show origin/main in /home/user/cloud and /home/user/hotcrm, read-only)",
  "api_writes": "This round 2 REST writes — PATCH /repos/objectstack-ai/objectstack/pulls/18480 ×1 (200; read-back: stored body = sent body + the platform's appended bare footer under a rule line, 2 footers total; the PATCH-channel behaviour AGENTS.md measures, ⛔ not re-sent); POST /repos/objectstack-ai/objectstack/issues/18477/comments ×1 (this report). Plus git push ×1 (365d1bc0). Cumulative on this card: POST /pulls ×1, POST /issues/18480/labels ×1, POST /issues/18477/comments ×2, PATCH /pulls/18480 ×1; git push ×3.",
  "deviations": "(1) The maintainer's sentence is quoted verbatim including 「不能让」; the ADR records the director's reading (「能让」) beside it. (2) Readings for J1, J2, J3, J6, J7, J10 are read-only REST reads of Actions runs / check-runs dated 2026-09-16, not from the named documents. (3) cloud runbook §7 / golden-journey.mjs kept ONLY as J3's runner, as the ruling allows; cloud `LOCAL-E2E-CHECKLIST.md` now appears only inside the 未入选 cloud lines. (4) `check:cross-package-test-inputs` :: 1 attributed to worktree build state, not to the diff (evidence above). (5) The PR body's second footer is the platform's, left in place per AGENTS.md. (6) Worktree recreated from the branch for this round and removed again after the push (tree clean, remote head 365d1bc0); no background process started.",
  "open_questions": [],
  "out_of_scope_findings": [
    "to file (a — reproducible; dedupe words: cross-package-test-inputs · spec/dist · init-created-files-summary · walked root · build-state): on origin/main 9c1897e5, `pnpm --filter @objectstack/spec build && pnpm check:cross-package-test-inputs` exits 1 naming packages/cli/test/init-created-files-summary.e2e.test.ts (REPO_ROOT = resolve(HERE, '../../..'), SPEC_PKG = resolve(REPO_ROOT, 'packages/spec')) walking into packages/spec/dist/ with no declared glob reaching it; with packages/spec/dist absent the same command exits 0 (measured both ways on this branch, whose code equals origin/main outside docs/adr). CI's Lint & Repo Gates is green on the same tip, so the verdict depends on whether a developer has built spec locally — which AGENTS.md instructs before judging anything. Two dispositions the gate itself prints: narrow the walk, or declare the root with a heldBy witness.",
    "noted, not filed (round 1, still true): the card names cloud ADR-0111 acceptance rows 1.1–5.x; the file on cloud origin/main (7be90d0) also carries R6 (6.1–6.2), so the honest count is 17, not 15. 承接者: the maintainer editing the table in PR #18480."
  ]
}

Generated by Claude Code

ACCEPT (round 2, after the maintainer's changes-requested review) — director seat, session_01Wj1HUjzyeiBQ8atRf1ZhaL, served tier claude-fable-5-1 · 2026-09-16T15:52Z

Verified against GitHub (PR #18480, head 365d1bc0, draft, 2 files +114/−0 vs origin/main; three-dot diff still exactly the two files), ⛔ not against the report — D1 read whole on pr/18480:

The table is now the developer platform's, as the maintainer's review 5224841038 directed: J1 discover + scaffold · J2 author metadata by hand · J3 AI-assisted authoring (cloud runbook §7 kept only as this row's runner) · J4 Studio zero-code loop · J5 os dev and see it in the console (objectui inside the row) · J6 verify, traps refused loudly · J7 package + publish · J8 install into an environment · J9 iterate against a customer requirement (hotcrm REQ line; none yet) · J10 the app's end users as the closing proof, one row only. Cap 10 met; 9 rows with a runner, 1 none yet; every newly cited runner checked to exist on its tree (git cat-file -e, per the report).
Cloud-service rows moved to 「未入选 — cloud 服务自己的清单」 (6 lines: signup→org→env, control-plane speed, hosted HotCRM login/space/isolation, operator enablement + blast radius, billing posture, cloud-side counter-sign) with the sentence that the cloud repo may declare its own list; 2 platform-side 未入选 lines. D2 (rulings, now J1–J10), D3, Consequences updated to the new sources. The maintainer's sentence is quoted verbatim with the director's 「能让」 reading beside it.
Gates: 18 derived / 18 run / 0 unrun (--ran, DERIVED zero); 17 exit 0; one red, not this PR's: check:cross-package-test-inputs exit 1 names packages/cli/test/init-created-files-summary.e2e.test.ts walking into packages/spec/dist/ — present only because the lock-wrapped formula/lint build built spec as a dependency; the same gate exited 0 on the same code before that build, and Lint & Repo Gates is green on origin/main tip 9c1897e5; filed as its own (a)-class card (below). mcp_calls 0; api_writes 2 this round, read back. CI on 365d1bc0 at review: 26 success · 11 skipped · 1 in progress · 0 failed.

Landing (governed docs/adr/**, ruling C): four-piece re-hung on PR #18480 — this record, 维护者速读(终稿 v2), needs-user-decision, review re-requested from os-zhuang (whose changes-requested review stands until re-read). ⚠️ The seat's ready-flip is refused by this session's harness classifier ([Merge Without Review], measured on this PR at 2026-09-16T15:34Z); the approver flips ready + enqueues after approving, as on #18476 / #18478 / #18481. Card stays pm:dispatched until MERGED.

Implemented-by: claude/issue-18477-declared-journeys-adr (mode:subagent)
Reviewed-by: session_01Wj1HUjzyeiBQ8atRf1ZhaL

Generated by Claude Code

Claim continues — round 3 (v2 → v3 rework) · director seat, session_01Wj1HUjzyeiBQ8atRf1ZhaL · 2026-09-17T07:28Z

Maintainer ruling on the direct channel, verbatim 「A(建议):」 after 「#18480(路的清单)有很大问题啊,基于元数据开发的应用到底能实现那些功能,开发好的应用是否要测试验证,这些需要进清单吗」 and 「这个清单的目的到底是什么」; recorded on PR #18480 (5710656572). Same branch claude/issue-18477-declared-journeys-adr, same card, a fresh mode:subagent os-dev round at the judgment tier (the round-2 subagent has ended; no SendMessage target). Worktree objectstack-issue-18477. Deliverable: v3 — main table 「应用能力 × 验证」 (one row per docs/qa/platform-checklist area, 15/15 covered), ≤6-row developer path, D2.4 「做好的应用必须过验证」; ⛔ no .claude/**, no scripts. Card stays pm:dispatched, assignee unchanged; four-piece re-hung on the v3 head after in-seat review. #18489 stays pm:blocked until acceptance.

Generated by Claude Code

os-dev-report

{
  "issue": 18477,
  "status": "done",
  "branch": "claude/issue-18477-declared-journeys-adr",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/18480",
  "session": "subagent of the director seat's session — harness-stamped id inherited (session_01Wj1HUjzyeiBQ8atRf1ZhaL)",
  "premise_still_valid": true,
  "summary": "v3 of docs/adr/0136-declared-journeys-as-priority-anchor.md, per the maintainer's 「A(建议)」 ruling. Head b3da1ec4; one file changed this round (+166/-59); three-dot vs origin/main is still exactly 2 files (+221): the ADR and round 1's single pointer line in PRIORITIZATION.md, which survived the merge of origin/main (e0d05538) and was NOT rewritten. TITLE: 「ADR-0136: 应用能力 × 验证,加开发者路径,作为优先级锚 — declared capabilities with their runners, and the ≤6-row developer path」. CONTEXT now opens with the purpose in one sentence: the record is the PM fleet's anchor for 先修什么, NOT a product-capability spec and NOT a test plan; the capability-and-verification ledger is docs/qa/platform-checklist/ (264 items / 15 areas, each with oracle and runner) and this table takes its rows from it. MAIN TABLE = 13 capability rows C1..C13 (C1 data modelling + record CRUD · C2 views + search · C3 forms/field types/validation/formula · C4 permissions, RLS, FLS, write-path guards · C5 approvals · C6 automation · C7 dashboards + reports · C8 identity + login · C9 integration + backend API · C10 attachments · C11 i18n · C12 AI · C13 Studio no-code), columns 应用里能做什么 / 声明它的元数据 / 怎么验证 / 最近读数 / 来源. SECOND TABLE = the developer path, 6 rows P1..P6 (enter · author metadata by hand/AI/Studio · run locally · verify · publish+install · iterate a customer requirement), folding v2's J1-J9. v2's J10 is deleted as a row and promoted to D2.4. AREAS 15/15 and ITEMS 264/264: a new 区 → 行 归属 table partitions every item — main table 244, path table 20, 未入选 0 — and names where platform-core's five P0s and cli's one P0 went. v2 cited 22 of 264; v3 cites all 264. Row cap is now BY CAPABILITY AREA (the hard ten is gone), the path table keeps a ≤6 cap. D2 is now four rulings: D2.1-D2.3 keep their wording with 旅程行号(J1-J10) replaced by 本表行号:C(n) 能力行 / P(n) 路径行 / none and 旅程断了或被挡 widened to cover a capability row or a path row; NEW D2.4 — an app built on the protocol counts as built only when (a) the checklist items its metadata kinds map to pass in a run record (selection via scripts/checklist-select.mjs, kinds via coverage.json metadataKinds, run-record shape per RUNNER.md) and (b) objectstack verify is green, with hotcrm as the reference implementation, and a card delivering app-facing behaviour with no passing run record is NOT done; D2.4 also states plainly that 110/264 items have an automated runner and 154 are manual, and names all 10 manual P0s, and that a manual verdict lands as a qa-run issue per runs/README.md. D3 unchanged. CONSEQUENCES updated: 清单即胃口 (cap by area; 12/13 capability rows have a runner, C11 is none yet; 5/6 path rows have a runner, P6 is none yet), 源与归并 (full recount, 264 items cited, 15/15 areas, and the correction that P0 is 20 not the 19 v2 recorded), 读数会腐烂 kept and extended with today's red Test Core, 执行是另一张卡 now says #18489 will carry the C/P line values, 代价 kept, plus a new line: a capability with no runner reads none yet and cannot feed the queue — that is the gap list for the checklist-author skill (C11 i18n · C5 approvals 2/18 · C13's P0 first-run-loop · P6 · the ten manual P0s). MOVED TO/FROM 未入选: the cloud block (6 lines) is kept verbatim; the platform-side block is EMPTIED because both its lines moved into capability rows — approvals.account-app-entry into C5 (C5's only P0) and studio-authoring.custom-page-render-and-blocks / custom-page-source-tiers into C13 — and the ADR says so explicitly. READINGS all re-taken 2026-09-17: Dogfood Regression Gate 🟢 on objectstack main@e0d05538 (CI run 35192708788); Test Core 🔴 on the SAME run (shard 2/6, packages/lint) so every row riding it reads unknown, never 🟢; scaffold-e2e #3088 🟢; check-links #9950 🟢 with the caveat that the workflow is pull_request-only and its last main run was a cancelled on 2026-08-30; platform-checklist-watchdog #14 🟢 on main@879b5127; showcase-smoke #90 🟢 2026-09-16 with #91 in_progress at read time; objectui live-e2e #4439 + ci #18088 🟢 on objectui main@e896c389 with the caveat that this is objectui main, not this repo's pinned .objectui-sha 53ded82b; cloud golden-journey #18 🟢 2026-09-16T18:33Z; hotcrm ci #3054 / e2e #1272 🟢 on hotcrm main@087b7c5d; hotcrm publish-staging #11 / publish-production #5 🟢 2026-09-14 on main@590b095e; hotcrm requirements 🔴 re-read at main@087b7c5d — all six REQ at Status: Triaged, REQ-0002 Traceability still 'to be filled in per phase', REQ-0003-0006 still 'to be filled in when built', plus the new fact that hotcrm draft PR #1950 (REQ-0006) is in flight with ci #3055 failure / e2e #1273 success.",
  "tests": "GATE DERIVATION: node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack (no paths passed; the script took the change set from the merge base itself — 2 paths, three-dot, merge base e0d05538c) → 18 commands, identical to rounds 1 and 2. Every one run in the FOREGROUND with the exit code captured by redirect then $? (never through a pipe). EXIT CODES: check-adr-links 0 (685 relative link destinations under docs/adr/ resolve) · check-adr-links --self-test 0 · check-adr-symbol-anchors 0 (2210 anchors across 140 records resolve) · check-adr-symbol-anchors --self-test 0 · check-ci-filter-parity 0 · check-closing-keyword-parity 0 · check-closing-keyword-parity --self-test 0 · check-comment-mask-corpus 0 (6803 files, 0 disagree) · pnpm check:adr-anchors 0 · pnpm check:nul-bytes 0 (8760 text files, no raw ASCII control bytes) · pnpm check:doc-authoring 0 · pnpm check:pm-governed-merges 0 · pnpm check:pm-prior-rulings 0 · pnpm check:driver-memory-census 0 · pnpm check:refd-timer-probe 0 · pnpm check:watch-hint-literal 0 · pnpm --filter @objectstack/lint run check:doc-formula-expressions 0 · pnpm check:cross-package-test-inputs 0 BEFORE the lint build and 1 AFTER it. RECONCILIATION: node scripts/pm/dispatch-gates.mjs --ran (with the cross-package line recorded as exit 1, the stricter of the two legs, so nothing is cherry-picked) → verbatim: '✓ dispatch-gates --ran: 18 derived famil(ies) accounted for — 18 run, 0 NOT-MEASURED (a DERIVED zero — all 18 recorded an exit code and none of them is 3).' VERIFY LOCK: read first with bash scripts/pm/os-verify-lock.sh --status (state: lock is free, queue empty). The only heavy step, building the @objectstack/lint dependency closure so check:doc-formula-expressions could run, went through the lock with OS_VERIFY_LOCK_SLOT=issue-18477-v3; its printed verdict line: 'os-verify-lock: VERDICT command-exit 0 · held the lock 183s (3m03s) · waited 0s'. THE CROSS-PACKAGE GATE, MEASURED BOTH WAYS ON ONE COMMIT: with packages/spec/dist absent (fresh worktree, docs-only diff) it exits 0; after building the lint closure produced that directory it exits 1, naming packages/cli/test/init-created-files-summary.e2e.test.ts descending into packages/spec/dist/ with 'no declared glob reaches inside it'. Same commit b3da1ec4, same diff, only the build state differs — this reproduces round 2's diagnosis as a controlled two-leg reading rather than restating it. This PR touches no test, no turbo.json and no package. SCOPE: no package is touched, so no dependency-closure build and no package test/typecheck is owed; repo-wide scans are CI's. NO ABLATION was owed — this round adds no gate and no guard, only ADR prose. ASSUMPTIONS: A1 VERIFIED — areas/*.json is 15 files / 264 items, per-area counts match the director's reading exactly (access-security 27 · ai 8 · api-backend 23 · approvals 18 · attachments-storage 9 · automation 16 · cli 17 · dashboards 11 · i18n 5 · identity-auth 22 · integration-system 18 · platform-core 29 · records-forms 39 · search 7 · studio-authoring 15) and all 264 are status:active — BUT its P0 half is FALSIFIED in the ADR's own favour: priority==='P0' counts 20, not the 19 v2's Consequences recorded (access-security 6 · platform-core 5 · api-backend 2 · ai 1 · approvals 1 · automation 1 · cli 1 · integration-system 1 · records-forms 1 · studio-authoring 1); the correction is written into Consequences and the checklist files are untouched. A2 VERIFIED — coverage.json carries metadataKinds keyed by LEDGER NAME, 37 entries, each {items:[...]} except realtime_subscription which is {waived:...}; the ADR's 声明它的元数据 column is declared as a READ of that map, so one kind may appear on several rows. A3 VERIFIED — scripts/checklist-select.mjs exists and prints its selectors when run with no argument: (id) | area: | capability: | priority:P0 | surface:api | since:vN | file:(path) | all; D2.4 quotes that set. A4 VERIFIED — all 13 runner paths v2 cites exist on the merged tree, 13/13 by git cat-file -e. A5 RE-READ, v2's verdict HOLDS — hotcrm main is still 087b7c5d (unmoved since 2026-09-16), all six REQ are Status: Triaged with placeholder Traceability; what v2 did not have and v3 now names is the in-flight hotcrm draft PR #1950 for REQ-0006 (ci #3055 failure / e2e #1273 success). SANITIZER DISCIPLINE: no HTML comments and no angle-bracket-shaped fragments anywhere in the PR body or this comment (scanned before sending); the ADR body was scanned for raw control bytes with grep -naP over the ASCII control classes — zero hits — and check:nul-bytes agrees. PR BODY READ BACK AFTER THE WRITE: the 13342 bytes sent survive BYTE-IDENTICAL (readback.startsWith(sent) true) and the platform APPENDED exactly the 58-byte block newline newline rule newline bare footer, leaving two footers — per AGENTS.md that tail bare footer is the platform's, so it was NOT re-sent and NOT repaired.",
  "mcp_calls": "0 — no MCP GitHub tool was called, read or write. All GitHub reads and both writes went through curl against the REST API with $GITHUB_TOKEN.",
  "api_writes": "2 — PATCH /repos/objectstack-ai/objectstack/pulls/18480 (body only; draft state, labels, reviewers and assignees untouched — read back after the write: state open, draft true, labels documentation / size/m / skip-changeset / needs-user-decision, unchanged) and POST /repos/objectstack-ai/objectstack/issues/18477/comments (this report). Plus one git push to claude/issue-18477-declared-journeys-adr (365d1bc0..b3da1ec4; the pre-push hook printed 'check:commit-card-trailers: 2 commit message(s) on this push carry no card relation and no model identifier in the trailer pair'). No label write, no POST /issues, no PATCH on any other card.",
  "files_changed": [
    "docs/adr/0136-declared-journeys-as-priority-anchor.md (rewritten, +166/-59 this round; 220 lines total)"
  ],
  "open_questions": [],
  "out_of_scope_findings": [
    "noted, not filed: objectstack main@e0d05538 is RED on Test Core shard 2/6 — packages/lint's `pnpm run test` exited 1 — in scheduled CI run 35192708788 at 2026-09-17T07:04Z, while Dogfood Regression Gate, Build Core, Temporal Conformance and Console Pin Gate on the same run are green. Not caused by this branch (docs-only), and it is the reason several ADR rows read `unknown` instead of green. 承接者: the next seat dispatched into packages/lint, or the main-red patrol. dedupe words: Test Core shard 2/6, packages/lint, main red, scheduled run 35192708788, 2026-09-17.",
    "noted, not filed: pnpm check:cross-package-test-inputs gives a BUILD-STATE-DEPENDENT verdict on an unchanged tree — exit 0 with packages/spec/dist absent, exit 1 with it present, measured both ways on commit b3da1ec4 with no diff between the legs. It flags packages/cli/test/init-created-files-summary.e2e.test.ts descending into packages/spec/dist/ ('no declared glob reaches inside it'). CI never sees it because CI builds before linting, so it only ever fires locally, on the agent that built a package closure to run some other gate. Round 2 reported the same shape; this round pins it with the two-leg control. 承接者: the gate's own owner — the fix is a declared glob for that walk, per the gate's own failure text. dedupe words: cross-package-test-inputs, init-created-files-summary.e2e.test.ts, packages/spec/dist, build-state-dependent, declared glob.",
    "noted, not filed: .github/workflows/check-links.yml has no `main` trigger — it runs on pull_request only, and its last run on the main branch is a `cancelled` from 2026-08-30 (run 33287387570). So the documentation-reachability runner the ADR cites for P1 has no reading on main at all; today's 🟢 (#9950) is from a PR run and the ADR says so in the cell. 承接者: whoever next edits that workflow or the docs-gate roster. dedupe words: check-links workflow, pull_request only, no main schedule, docs link reachability.",
    "noted, not filed: docs/adr/PRIORITIZATION.md's pointer line names ADR-0136 by its v1 title ('Declared journeys as the priority anchor'), which is no longer the H1. check-adr-links passes (it resolves destinations, not link text) and this dispatch fences that file, so it was left byte-unchanged. 承接者: whoever lands the enforcement card #18489, which touches this area next. dedupe words: PRIORITIZATION pointer, ADR-0136 title drift, link text."
  ]
}

Generated by Claude Code

ACCEPT (round 3, v3 after the maintainer's ruling A) — director seat, session_01Wj1HUjzyeiBQ8atRf1ZhaL, served tier claude-fable-5-1 · 2026-09-17T07:56Z

Verified against GitHub and the branch tree (PR #18480, head b3da1ec4, draft; three-dot vs origin/main = exactly 2 files, +221: the ADR (220 lines) and round 1's one pointer line in PRIORITIZATION.md, unchanged), ⛔ not against the report:

Shape = ruling A. Title re-keyed; Context opens with the purpose sentence (anchor for 先修什么, not a spec, not a test plan; the checklist is the ledger). Main table C1–C13 (data+CRUD · views+search · forms/fields/formula · permissions · approvals · automation · dashboards · identity · integration+API · attachments · i18n · AI · Studio), columns 应用里能做什么 / 声明它的元数据 / 怎么验证 / 最近读数 / 来源. Path table P1–P6 (enter · author by hand/AI/Studio · run locally · verify · publish+install · iterate a customer requirement). v2's J10 became D2.4. D2.1–D2.3 keep their wording with C<n>/P<n>/none; D3 unchanged.
Arithmetic re-done by the seat on the branch tree: 15 areas / 264 items, per-area counts match; the 区 → 行 table partitions all 264 (main 244 + path 20, every area row sums to its file: cli 3+1+2+5+5+1, platform-core 9+12+1+7, records-forms 14+10+14+1, api-backend 4+19); main-table row item counts sum to 244. automated items 110 / manual 154; P0 = 20 (v2's 19 was wrong — the correction is in Consequences, checklist files untouched); the 10 manual P0 ids the ADR names are exactly the 10 the tree yields; coverage.json metadataKinds 37 with 1 WAIVED; checklist-select.mjs prints the 8 selectors D2.4 quotes.
Readings all dated 2026-09-17 with run links; Test Core red on main@e0d05538 (shard 2/6, packages/lint) is carried as unknown on every TC-dependent row, ⛔ not painted green; objectui reading is stated as objectui main, not the pinned .objectui-sha; check-links is stated as a PR-only run. Two rows read none yet (C11 i18n, P6 customer-requirement iteration) and the Consequences turn them into the gap list.
未入选: cloud block kept verbatim (6 lines); platform-side block emptied with both lines' destinations named (C5, C13).
Gates: 18 derived / 18 run / 0 unrun (--ran, DERIVED zero); 17 exit 0; check:cross-package-test-inputs exit 0 before the lint build and 1 after it on the same commit — the objectstack#18495 shape, now pinned with a two-leg control; ⛔ not this PR's. CI on b3da1ec4 at this read: 24 success · 11 skipped · 2 in progress · 0 failed. mcp_calls 0; api_writes 2 (PR body PATCH, report) + 1 push, read back; no label/reviewer/assignee touched by the dev.
Out-of-scope lines: main Test Core red (scheduled run 35192708788) — the main-red patrol's, not carded here; check-links.yml has no main trigger — noted for the workflow's next editor; PRIORITIZATION.md pointer text names the v1 title — rides skills(pm-dispatch): ADR-0136 D2 becomes the triage protocol — Journey: line, journey bands, cross-lane product-first order, journeys feed the queue #18489.

Landing (governed docs/adr/**, ruling C): four-piece re-hung on PR #18480 — this record, 维护者速读(终稿 v3), needs-user-decision (already on the PR), review re-requested from os-zhuang (whose v1 changes-requested review is the latest decisive one, so the queue guard needs a fresh APPROVED on b3da1ec4). The seat's ready-flip is refused by this session's classifier; the approver flips ready + enqueues. Card stays pm:dispatched until MERGED; #18489 stays pm:blocked until then.

Implemented-by: claude/issue-18477-declared-journeys-adr (mode:subagent, round 3)
Reviewed-by: session_01Wj1HUjzyeiBQ8atRf1ZhaL

Generated by Claude Code

Closed — superseded by the North Star (skills seat, session_01BTeBejoPUvRHN8WdAJC6oF, on the maintainer's word) · 2026-09-18T22:09Z

PR #18480 (ADR-0136 v3) is closed unmerged at the maintainer's decision (chat 2026-09-18, 「同意,北极星定稿 … 136是否要关闭」 → 「其他同意」; the record is on the PR). What this card asked for — a priority anchor for the fleet — is answered by docs/NORTH-STAR.md (the maintainer's text, finalized in chat, landing through its own card) and enforced by #18489 (re-scoped to 「按北极星执法」). pm:dispatched removed, assignee cleared; closed not_planned as superseded, ⛔ not as rejected work — the readings in the ADR drafts informed the North Star.

Generated by Claude Code

Navigation Menu

Repository navigation

docs(adr): declared journeys as the priority anchor — the maintainer's ≤10 end-to-end paths, each with its runner, become the first triage question (maintainer 「立」) #18477

Why (director's readings, 2026-09-16)

Deliverable — ONE ADR, draft, the maintainer edits by deletion

The journey table — ≤10 rows, drafted from existing sources, ⛔ nothing invented

The decision section — three rulings, stated as rules the PM protocol will cite

Consequences section

Acceptance

What the rework changes (D1 only; D2 rulings and D3 supersession stand)

Execution

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

docs(adr): declared journeys as the priority anchor — the maintainer's ≤10 end-to-end paths, each with its runner, become the first triage question (maintainer 「立」) #18477

Description

Why (director's readings, 2026-09-16)

Deliverable — ONE ADR, draft, the maintainer edits by deletion

The journey table — ≤10 rows, drafted from existing sources, ⛔ nothing invented

The decision section — three rulings, stated as rules the PM protocol will cite

Consequences section

Acceptance

Activity

hotlong commented on Sep 16, 2026

hotlong commented on Sep 16, 2026

hotlong commented on Sep 16, 2026

hotlong commented on Sep 16, 2026

What the rework changes (D1 only; D2 rulings and D3 supersession stand)

Execution

hotlong commented on Sep 16, 2026

hotlong commented on Sep 16, 2026

hotlong commented on Sep 17, 2026

hotlong commented on Sep 17, 2026

hotlong commented on Sep 17, 2026

os-elon-musk commented on Sep 18, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions