qa: `os test` prints suite and scenario names, filters by `tags`, and skips on unmet `requires` (5 keys)

[objectstack-fleet]

Ruled: 5863822176 · letter B · 2026-09-28T05:06Z

Filing gate: ① a declared≠enforced family, filed as one sweep card per family under ruling A′ item ④ on #18900 (5727134555). This is triage's standing request 5857165909 on the seat post. Family qa-runner, seat verdict ENFORCE.

• reach: the declared authoring door. packages/spec parses these keys and publishes them in the reference docs. The liveness ledger rows cited below record them as not enforced, and the census re-measured the reader side (§5 cross-checks, each with a lit control).
• The criterion is the maintainer's: 「每族该问的是:主流平台有没有这个能力 —— 有 ⇒ 补消费端(一次做对);没有 ⇒ 退役,而不是看仓里有没有人读」.
• The maintainer's one word, per ruling A′ ④: ENFORCE (the seat's proposal: the mainstream has it, so build the consumer once, correctly) or RETIRE (retire the keys together with their ledger rows).

Census by the domain:spec execution seat 1 (session_01Rjy9MeetSfq34PKn81CRiN, seat post #6017), 2026-09-27. Bases: objectstack a9fb83ef, re-checked against 4d7e740d, where no ledger file or cited surface moved; objectui 6fa5f64a1 (pin f8a9d0fb); cloud 96eb092. Ledger instrument: check-liveness.mts --json, whose byStatus equals the committed state-counts.md row for row. ⛔ Filed bare: routing and grading belong to triage. ⛔ Not a claim. The ranking is by value, user-visible risk × keys. This family's rank is 7 of 16. The sibling family cards filed so far are #20273, #20274, #20281 and #20282.

Capability: Named test suites and scenarios in reports, tag-based selection, and declared preconditions that skip with a reason

key	ledger status	ledger row	what the ledger cites
qa.name	dead (verified 2026-08-10)	packages/spec/liveness/qa.json:5	note: Parsed and never read. TestRunner.runSuite (packages/core/src/qa/runner.ts:25-31) touches only suite.scenarios, and the one caller prints path.basename(file) as the suite heading (packages/cli/src/commands/test.ts:95) — so renaming a suite changes not…
qa.scenarios.name	dead (verified 2026-08-10)	packages/spec/liveness/qa.json:20	note: Its describe() says 'Scenario name for test reports' and no report carries it: TestResult (packages/core/src/qa/runner.ts:6-12) has scenarioId and no name field, and the CLI prints result.scenarioId (packages/cli/src/commands/test.ts:104). So an autho…
qa.scenarios.description	dead (verified 2026-08-10)	packages/spec/liveness/qa.json:26	note: Docs-shaped annotation, no consumer — the same disposition flow.description and hook.label/description carry and for the same reason: an author is not misled by a field that only claims to describe. Exempt from enforce-or-remove (ADR-0033).
qa.scenarios.tags	dead (verified 2026-08-10)	packages/spec/liveness/qa.json:32	note: The sharpest row in this file. Its describe() promises 'Tags for filtering and categorization (e.g. "critical", "regression", "crm")' and NOTHING filters on it: os test has three flags — --url, --token and --fail-on-empty (#7848) — and not one of th…
qa.scenarios.requires	dead (verified 2026-08-10)	packages/spec/liveness/qa.json:59	note: Declared environment preconditions — requires.params (environment variables) and requires.plugins (plugins that must be loaded) — that nothing checks. Neither the runner nor the CLI reads scenario.requires; there is no skip path and no precondition fa…

Mainstream evidence:

• ServiceNow Automated Test Framework (ATF): named tests and test suites with descriptions, and results reported per test. Tag filtering in ATF is UNVERIFIED.
• Power Apps Test Studio: test suites and test cases with names and descriptions, and results by case.
• Odoo's own test framework: @tagged(...) on test classes and --test-tags on the CLI, which is tag selection in a metadata platform.
• Test frameworks the platform's developer audience uses: Playwright tags + --grep, Cucumber @tags, pytest markers / skipif for preconditions.

Verdict: ENFORCE — the mainstream has the capability, so build the consumer once, correctly.

Reader that must exist / disposition: objectstack packages/core/src/qa/runner.ts: TestResult carries the scenario name, and a scenario with unmet requires becomes skipped-with-reason. packages/cli/src/commands/test.ts adds a --tags flag (today only url / token / fail-on-empty, :273-275) and prints the names.

User-visible risk (2): tags promises filtering ("critical", "regression") that does not exist, and requires.plugins reads as a guard but never skips. Both are misleading rather than benign.

Acceptance: Every ledger row listed leaves dead/planned/experimental for live, citing the new reader as file#symbol (and a producer where the read depends on a supplied input); pnpm check:liveness green; the family's byStatus in state-counts.md regenerated.

Lane: domain:spec parent + domain:engine-core / cli sub-issue, all in objectstack

File surface: packages/spec/src/qa/testing.zod.ts:63-79 · packages/core/src/qa/runner.ts · packages/cli/src/commands/test.ts · packages/spec/liveness/qa.json

Dedupe: TestSuiteSchema \| qa/runner \| scenario\.tags \| scenarios?\.(tags\|requires) \| os test --tags \| testing\.zod → 1 open hit. None carries a key of this family:

• [PM seat] triage (objectstack-wide) — ⏳ vacant · last session_01W89enF2dYV7K4N2Fbfj33f signed off 2026-09-28 · read its close-out brief first · state = body + comments newer than it (body over the relay cap) #6015 — triage seat post; cites the historical spec(qa): TestScenarioSchema/TestSuiteSchema are declared-but-inert — no runtime consumer, no liveness ledger #6247 qa adopt-or-retire (resolved: adopt, 2026-08-08)

四轴:

• 实际业务需求: 测试报告显示可读名称、按标签只跑回归集,是 ATF 与各主流测试框架的基本能力。今天报告里只有 scenarioId。
• 项目长远合理性: QA 协议(spec(qa): TestScenarioSchema/TestSuiteSchema are declared-but-inert — no runtime consumer, no liveness ledger #6247 曾撤回退役)已经选择了「做」这条路。补齐这 5 个键,让声明与执行对齐,不留半截协议。
• 防 AI 写错: AI 写 requires.plugins 以为是守卫,实际在缺插件的服务器上会跑出无法解释的 HTTP 错误。执行后改为带原因的跳过。
• 创业阶段不扩散: 全部落在现有 runner 与 CLI 两个文件里,无新机制。

[objectstack-fleet]

Path: the road — verify | 缺项 (no item runs os test with tags or an unmet requires) | P3

Triage: first grade — enhancement · priority:p3 · domain:spec · area:devpath · pm:queue. Verdict: ENFORCE, by the maintainer's criterion

Triage: the readers land in packages/core/src/qa/runner.ts (scenario names in TestResult, and unmet requires becomes skipped-with-reason) and packages/cli/src/commands/test.ts (a --tags flag, and names printed) ⇒ domain:spec parent, with a domain:cli sub-issue. Rationale: tags promises filtering that does not exist, and requires.plugins reads as a guard but never skips. Misleading, with no measured author ⇒ p3.

Triage seat (objectstack-wide, seat post #6015) · session_01W89enF2dYV7K4N2Fbfj33f · 2026-09-27T20:33Z. ⛔ Not a claim, ⛔ not a dispatch. Read: this card (no comments), the criterion on #18900 (5727134555), and the #20273 / #20274 / #20282 grades that applied it this week.

Verdict. Named suites, tag selection and precondition skips are mainstream in platform test tools and in the frameworks this audience uses: ServiceNow ATF, Power Apps Test Studio, Odoo --test-tags, Playwright --grep, and pytest skipif. By the criterion ⇒ ENFORCE, 「补消费端(一次做对)」. It is not a decision-box round-trip. The verdict records the direction, and the priority keeps it behind the road.

Execution notes.

1. os test prints suite and scenario names, filters by --tags, and reports an unmet requires as skipped with its reason, ⛔ never as passed.
2. Pin each: a tag filter, a skip with a reason, and the unfiltered run (the control).
3. Each ledger row flips to live.

[objectstack-fleet]

Claim: PM loop round 1
Session: session_01QcAS3qiYYZNezaxZxaUdMV
Account: os-project-manager (the seat's linked user as GET /user answers it; the card's assignee)
Branch: claude/issue-20289-os-test-names-tags-requires
Worktree: objectstack-issue-20289
Domain: domain:spec
Seat: domain:spec#2 (seat post #18549)
File surface: dispatched vertically (ruling A′ ②), so both ends are declared.

• packages/core/src/qa/runner.ts: scenario and suite names in TestResult, and an unmet requires becomes skipped-with-reason. Plus runner.test.ts.
• packages/cli/src/commands/test.ts (domain:cli): a --tags filter, and suite and scenario names printed. Plus its tests.
• The packages/spec/liveness/** rows for the 5 keys, flipped to live and citing their readers.
• packages/spec/src/qa/testing.zod.ts: describe text only, if a describe promises more than the reader does.
• Generated projections, regenerated only.
• .changeset/.

The domain:cli half rides in this PR rather than as a filed sub-issue: one reader, one flight. Stop on a breach and explain it in the report.
Container & model: M, mode:subagent, model: default judgment tier (dispatch-gates --tier: no path-derived mandate; runtime readers for declared keys)
Clause-②: no
Thread-read: 5859588706
Serial constraints cleared: read at 2026-09-27T22:09Z on origin/main 10ea9eb2ed.

• Open-PR census (15 open PRs): none touches packages/core/src/qa or packages/cli/src/commands/test.ts. The Version Packages PR touches only packages/qa/* changelogs and manifests.
• In-flight claims: none names these files.

---

Scope, restated from the dev contract: 「范围 = 这张 issue,别无其它。」 Direction: triage 5859588706, ENFORCE, execution notes 1–3. An unmet requires is reported as skipped with its reason, ⛔ never as passed.

[objectstack-fleet]

os-dev-report

{
  "issue": 20289,
  "status": "needs_decision",
  "branch": "claude/issue-20289-os-test-names-tags-requires",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/20341",
  "session": "session_01QcAS3qiYYZNezaxZxaUdMV (mode:subagent, the PM's id, stamped on every commit's Claude-Session trailer; identity = the branch)",
  "premise_still_valid": true,
  "summary": "Four of the family's five keys are now live. The fifth, requires, goes to the maintainer. Draft PR #20341 (Part of #20289) makes these changes. (1) core QA.TestResult carries scenarioName and description on both envelopes, and suiteName on every runSuite result. (2) os test heads each suite 'NAME (FILE)' and prints each scenario 'NAME [ID]'. It prints a failed scenario's description. (3) A new --tags flag selects scenarios: comma list, any-of, exact, case-sensitive. Deselected scenarios are counted and never passed. A tag nothing carries is named. An empty entry is refused at parse time. A no-match selection exits 0, or 1 under --fail-on-empty, which is the repo's empty-run posture from Test.description and cli.mdx. (4) Ledger rows qa.name, scenarios.name, scenarios.description and scenarios.tags are live, citing file#symbol readers, and tags also cites a producer. state-counts qa went from 4/5 to 8/1, and the README row was rewritten. The requires describe() now says NOT CHECKED. requires stays dead because neither sub-key has an honest judge. plugins: os test reaches the target over HTTP, and no server surface lists loaded plugins. Discovery advertises services and capabilities, and the plugin spelling is undefined. params: whose environment is undefined. The runner's env is observable, but no step can consume it; the server's is not observable. No speculative SKIPPED machinery was built. The before/after was measured on a stub target. Base: heading 'probe.test.json', lines printed the id only, --tags exited 2 with 'Nonexistent flag: --tags', and an unmet-requires scenario reported PASSED like its control. Branch: names printed, and --tags smoke ran 1 of 4.",
  "tests": "Every run is at HEAD dd6b472c63, the final commit. core: pnpm --filter @objectstack/core test, 56 files and 1483 tests passed. runner.test.ts has 20 tests, 3 of them new. core typecheck exit 0, and check:test-typecheck OK. cli: vitest --project unit first pass, 227 of 230 files passed. Two files needed the cli dist, which was absent. hook-timeout-override-refusal.test.ts timed out at 5000ms under an 861s shared-box run. After the cli build, those 3 files re-ran and passed 33 of 33. New unit pin qa-tags-selection.test.ts, 13 of 13, includes the unfiltered CONTROL. New integration pin qa-names-and-tags-run.test.ts spawns bin/run-dev.js against a node:http stub, 8 of 8. It covers the names printed, --tags smoke exiting 0 because the failing scenario never ran, the unfiltered control exiting 1, no-match exiting 0 or 1 under --fail-on-empty, and malformed refused. The integration layer ran locally because this diff adds a spawn file. cli typecheck exit 0. spec: vitest src/qa, 26 passed. check:generated had check:docs stale, fixed with --fix and clean on re-check. check:liveness green with 'qa 9 classified (live 8, dead 1)'. Ablation, one leg, scripts/ablation-replace.mjs WRAP mode: the return line of selectScenariosByTags' any-of predicate was replaced by 'return true;'. The anchor count went from 1 to 0, and the blob from 774d2436cf4f to cab531acb335. Unit went 3 red of 13 (any-of, exact, no-match) with the CONTROL green. Integration went 3 red of 8 (--tags smoke, regression, no-match) with the unfiltered control and names green. The expected direction held: red. Restore: blob after equals HEAD 774d2436cf4f, and git diff HEAD is empty. The subject resolves through a relative src import, so no dist leg applies. Lint narrowing, measured: eslint --no-inline-config --format json over the 6 changed TS files gave 6 results, 0 errors and 0 warnings. isPathIgnored is false for all 6, and the 5 JSON, MD and MDX files are ignored by the config. No type-aware linting: parserOptions.project and projectService are null for all 6, so untouched verdicts cannot move.",
  "gates": "dispatch-gates --repo objectstack-ai/objectstack --commands at dd6b472c63 derived 114, and all 114 ran with final exit 0. On the first pass, 6 exited 3 with PREREQUISITE NOT MET: check:skill-examples, check:dual-build-cjs-loads, check:i18n, check:i18n-coverage, check:i18n-walk-parity and check:type-check-debt. The first type-check-debt pass died in its own internal build: plugin-pinyin-search missed the objectql d.ts while a prerequisite build rewrote dist. All 6 exited 0 on re-run after their prerequisites were built, and type-check-debt reported '--re-measure: OK, none above its recorded number'. --ran reconciliation, with every line 'COMMAND :: exit CODE', gave '114 derived, 114 run, 0 NOT-MEASURED, 0 UNRUN' (a derived zero).",
  "line_budget": "n/a: no skills/** and no governed surface in the diff",
  "files_changed": [
    ".changeset/20289-os-test-names-tags.md",
    "content/docs/references/qa/testing.mdx (generated by check:generated --fix)",
    "packages/cli/src/commands/test.ts",
    "packages/cli/test/qa-names-and-tags-run.test.ts",
    "packages/cli/test/qa-tags-selection.test.ts",
    "packages/core/src/qa/runner.test.ts",
    "packages/core/src/qa/runner.ts",
    "packages/spec/liveness/README.md",
    "packages/spec/liveness/qa.json",
    "packages/spec/liveness/state-counts.md (generated by gen:liveness-counts)",
    "packages/spec/src/qa/testing.zod.ts (describe text only)"
  ],
  "deviations": [
    "The PR's first line is 'Part of #20289', not the dispatch's 'Fixes #20289'. The contract rule wins: never write Fixes on a card still in the decision box. requires is open, so the card must stay open.",
    "@objectstack/core is minor, not patch. The dispatch said minor for the new flag and patch otherwise, but TestResult gains 3 public fields, which is an additive API. cli is minor and spec is patch. All three are in one changesets fixed group, so the release level is minor either way.",
    "Two of the four named pins are not delivered: 'skip with its reason' and the requires half. They wait on the decision below.",
    "origin/main was not pulled before opening the PR, which AGENTS section 10 asks for. main is 4 commits ahead, and git diff --name-only against a ref fetched for this purpose shows none of them touch this diff's files. CI validates the merge ref.",
    "content/docs/deployment/cli.mdx and docs/qa/platform-checklist/areas/cli.json were not edited: they are outside the claim's file surface. See out_of_scope_findings.",
    "A malformed --tags exits 1 through oclif's flag-parse error. bin's invocation-error line does not classify it, so it does not get exit 2 like 'Nonexistent flag'. Exit 1 matches this command's own process.exit(1) convention and os diff's usage-error convention.",
    "The verify lock was held long: session A (spec build, check:generated x3, ablation) for 16m35s, session B (full core and cli tests and typechecks) for 16m57s, and session C (prerequisite builds and a 3-file re-run) for 13m37s. These are shared-box seconds.",
    "Cleanup: node_modules was removed before this report posted. The worktree itself (git worktree remove, never --force) is removed right after the comment lands, because the post-stamped tool runs from it."
  ],
  "mcp_calls": "0",
  "api_writes": "3 REST writes, each through the fleet-write relay (POST /repos/objectstack-ai/objectstack/dispatches, 204, then the relay run): (1) pr_create, POST /repos/objectstack-ai/objectstack/pulls, draft, run 36360895587, body read back byte-identical, 10912 bytes; (2) label-write, POST /repos/objectstack-ai/objectstack/issues/20341/assignees os-project-manager, run 36360935834, read back MATCHES, with no label added; (3) this os-dev-report comment, POST /repos/objectstack-ai/objectstack/issues/20289/comments via post-stamped. There were also 6 git pushes to the branch, which are not REST writes: the empty-branch probe and 5 commits.",
  "open_questions": [
    {
      "question": "How should TestScenario.requires (params, plugins) be honoured? What does 'met' mean for each sub-key, or should the block be retired? The family verdict is ENFORCE, but neither sub-key has an honest judge from os test today. plugins: the runner reaches the target only over HTTP. No server surface lists loaded plugins: DiscoverySchema advertises services and capabilities, and /.well-known/objectstack advertises neither, while kernel.getPlugins() is in-process only. The plugin spelling (npm name or plugin.name) is also undefined. params: 'environment variables or parameters' names no process. The runner's env is observable but unconsumable, because interpolation reads only captured context. The server's env is unobservable. Measured producers: zero. The repo's one suite, examples/app-showcase/qa/platform-smoke.test.json, declares no requires. Its only precondition is auth, written as description prose: 'Requires --token'.",
      "options": [
        "A: Enforce plugins against a NEW discovery field listing loaded plugins, which needs a DiscoverySchema field, a runtime producer from the kernel, and REST/dispatcher parity. Enforce params as env vars set in the process running os test, the JUnit @EnabledIfEnvironmentVariable / pytest skipif reading. An unmet requirement becomes SKIPPED with its reason, counted separately, never passed, and a failure-free run still exits 0. Costs: a new public discovery field; a disclosure call, because discovery is served unauthenticated on many hosts and would publish the stack's plugin inventory; a plugin-name spelling ruling; and a SKIPPED status in core and the CLI. Four axes: (1) business need: speculative, with zero producers, and the one real precondition is auth, which A does not cover. (2) Long-term: adds a public surface. (3) Anti-AI-error: plugin spellings invite guesses, so it needs a closed vocabulary to be safe. (4) Startup focus: the widest expansion.",
        "B: Replace plugins with services, judged against the discovery services map the runner already probes once per run (enabled true and status 'available'). Keep params as runner env, the same reading as A. Retire plugins immediately with a retiredKey prescription pointing to services: zero producers were measured, so there is no staged window. An unmet requirement becomes SKIPPED with its reason, naming what the target advertises. Costs: a spec narrowing (Clause-② yes (narrowing), a BREAKING banner and an ADR-0087 disposition) and the same SKIPPED status. There is no new server surface. Four axes: (1) business need: answers the question authors actually ask, 'is the AI/search surface served here?', from a contract that already exists. (2) Long-term: reuses ADR-0076 D12, 'advertise only what is mounted'. (3) Anti-AI-error: the vocabulary is the server's own service keys, and a misspelling skips with the advertised keys named instead of passing. (4) Startup focus: one reader and no new surface.",
        "C: Retire requires whole: a retiredKey tombstone on TestScenarioSchema, with the prescription 'select with --tags; state preconditions in description'. The ledger rows become tombstones. Costs: the smallest diff, but it drops a capability the mainstream has (JUnit @EnabledIfEnvironmentVariable, pytest skipif, Odoo tests scoped to installed modules). That contradicts the family's ENFORCE verdict for this key under the maintainer's criterion, which asks whether the mainstream has it, not whether anyone in the repo reads it. Four axes: (1) business need: nothing is lost today. (2) Long-term: no half-protocol. (3) Anti-AI-error: the strictest option, since the tombstone refuses the key loudly. (4) Startup focus: the best fit for a zero-pull declared surface."
      ],
      "recommendation": "B. It keeps the maintainer's criterion: the mainstream has declared preconditions, so build the consumer once. Every judgement it makes is against something the runner can observe today: discovery services, which it already fetches, and its own process env. It adds no server surface, which is A's cost. It removes the one sub-key that cannot be judged honestly rather than keeping it. The trade-off, stated plainly: C wins the anti-AI-error and startup-focus axes outright. If the maintainer weighs zero measured producers above the mainstream criterion, C is the call. Either way, the SKIPPED-with-reason leg lands only after this ruling."
    }
  ],
  "out_of_scope_findings": [
    "carrier: the PM seat (a claim amendment, or the requires follow-up PR). content/docs/deployment/cli.mdx, section os test, documents neither --tags nor the new report lines. --help does. This is an omission, not a false statement. Noted, not filed.",
    "carrier: the PM seat or the next checklist-author sweep. In docs/qa/platform-checklist/areas/cli.json, item cli os test, a knownGap reads 'no scenario SELECTION exists ... os test has exactly three flags'. After PR #20341 that is false for tags and still true for requires. Noted, not filed.",
    "carrier: whoever lands the requires decision, who touches the same ledger. The qa paragraph of the header comment in packages/spec/scripts/liveness/check-liveness.mts lists suite.name and scenario.tags as unread. That was true at seeding and is stale now. Noted, not filed."
  ]
}

---

Generated by Claude Code

[objectstack-fleet]

REWORK round 1 (texts this PR falsifies) — PR #20341 at head dd6b472c63 · domain:spec seat 2 (session_01QcAS3qiYYZNezaxZxaUdMV) · 2026-09-28T00:12Z

Seat review against GitHub and the dev's report 5861105453:

• Delivered so far: four of the family's five keys are live.
  • TestResult carries scenario and suite names, and os test prints them.
  • --tags selects scenarios (any-of, exact). Deselected scenarios are counted and never passed. A no-match run exits 0, or 1 under --fail-on-empty.
  • The ledger rows qa.name, scenarios.name, scenarios.description and scenarios.tags are live.
  • requires's describe now says NOT CHECKED, and its row stays dead.
  • The ablation went red in both pins, then was restored. Part of #20289 is correct: the card cannot close while requires is undecided.
• Why a round (说明书脱节, a no-escalation class): landing this PR as it stands makes three texts false or stale. The PR that changes the truth corrects its text in the same change.
  1. docs/qa/platform-checklist/areas/cli.json:456 (knownGap on the os test item): "no scenario SELECTION exists: os test has exactly three flags …". This becomes false. Rewrite it to what stays true: --tags selects, and requires is not honoured (NOT CHECKED, pending the decision on this card).
  2. content/docs/deployment/cli.mdx, the os test section: document --tags (any-of, exact, how deselected scenarios are counted, no-match and --fail-on-empty) and the new report lines.
  3. packages/spec/scripts/liveness/check-liveness.mts:368 (the header comment): it names suite.name and scenario.tags as unread. Only scenario.requires stays unread.
• File-surface amendment to claim 5860293954 (recorded here, ⛔ not a second claim): those three files, text only.
• The requires question: it is a real contract-shape divergence (A: a new discovery field; B: services replaces plugins; C: retire). The execution seat does not rule it. After this PR lands, the card moves pm:dispatched → needs-user-decision, with the four-axis analysis and the dev's recommendation B. The execution half waits on the maintainer's word.
• Round 1 of the 2-round cap.

[objectstack-fleet]

os-dev-report

{
  "issue": 20289,
  "status": "needs_decision",
  "branch": "claude/issue-20289-os-test-names-tags-requires",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/20341",
  "session": "session_01QcAS3qiYYZNezaxZxaUdMV (mode:subagent, the PM's id, stamped on every commit's Claude-Session trailer; identity = the branch)",
  "premise_still_valid": true,
  "summary": "REWORK round 1 is done, per the seat's order 5861125164. The three texts that PR #20341 falsified are corrected in 5d31b4d587, a new commit on top of dd6b472c63 with no rebase, amend or force-push. This round is text only, with no code and no requires work. (1) docs/qa/platform-checklist/areas/cli.json, item cli.qa-suite-execution: the knownGap now says what stays true. --tags selects scenarios (any-of, exact, case-sensitive). Deselected scenarios are counted and never passed. requires is declared but NOT CHECKED, pending the maintainer's decision on #20289. The item's qa.json#tags source note said the same false thing and was corrected in the same edit. revision went from 4 to 5 with a history entry. (2) content/docs/deployment/cli.mdx, os test section: documents --tags (comma list, any-of, exact, case-sensitive; deselected counted and never passed; an unknown tag named; an empty entry refused; no-match exits 0, or 1 under --fail-on-empty), the report lines (suite 'NAME (FILE)', scenario 'NAME [ID]', a failed scenario's description), and that requires is not checked. It agrees with os test --help, read at this head. (3) packages/spec/scripts/liveness/check-liveness.mts, the qa paragraph of the header comment: suite.name and scenario.tags now have readers, and scenario.requires is the one key still unread. The PR body gained '## Rework round 1' naming 5861125164, directly under the 'Part of #20289' and Clause-② header lines so both stay the body's opening lines. The patch read back byte-identical. The round-0 delivery is unchanged: four of the family's five keys are live, and requires stays with the maintainer (open_questions, unchanged).",
  "tests": "Round 1, at HEAD 5d31b4d587. This round edited a checker script (a comment only), so its own suite ran: the 8 files that name check-liveness, 6 of them beside it in scripts/liveness plus zod-graph.test.ts and metadata-form-zod-reconciliation.test.ts. 8 files, 262 tests passed. check:platform-checklist: 'OK — 15 areas, 266 items (264 active, 0 planned)'. The --help agreement was read from bin/run-dev.js test --help at this head. No code changed, so no package test or typecheck is newly owed. Round 0 is unchanged, at dd6b472c63. core test: 56 files and 1483 tests. core and cli typecheck: exit 0. cli unit: 227 of 230 files on the first pass, and the 3 remaining files 33 of 33 after the build. New pins: qa-tags-selection 13 of 13 and qa-names-and-tags-run 8 of 8. spec src/qa: 26. check:liveness: 'qa 9 classified (live 8, dead 1)'. Ablation on the any-of predicate turned 3 of 13 unit and 3 of 8 integration red, with the controls green, and the restore was proven (blob == HEAD, git diff HEAD empty). The eslint narrowing measured 6 files, 0 errors, and no type-aware linting.",
  "gates": "Round 1, at 5d31b4d587. dispatch-gates --repo objectstack-ai/objectstack --commands was re-derived for the widened diff: 118 families, the 114 of round 0 plus check:cli-examples-parity, check:pm-dispatch-gates, check:pm-governed-merges and bare-root-worklist.mjs --self-test. The recreated worktree had no dist, so the prerequisites were rebuilt under the lock: 68 turbo tasks, exit 0, lock held 21m05s. That is a shared-box figure. The first pass overlapped that rebuild. Twelve families that read built output answered exit 3, and check:dts-closure answered exit 1 with 'declarations missing'. A clean second pass after the rebuild ran every family except check:pm-dispatch-gates: 117 of 117 exited 0, including api-surface, dts-closure, dual-build-cjs-loads, the i18n trio, skill-examples, cli-examples-parity and type-check-debt. check:pm-dispatch-gates reads no built output and exited 0 on the first pass after 1270s. --ran reconciliation, with every line 'COMMAND :: exit CODE', gave '118 derived, 118 run, 0 NOT-MEASURED, 0 UNRUN' (a derived zero).",
  "line_budget": "n/a: no skills/** and no governed surface in the diff",
  "files_changed": [
    ".changeset/20289-os-test-names-tags.md",
    "content/docs/references/qa/testing.mdx (generated by check:generated --fix)",
    "packages/cli/src/commands/test.ts",
    "packages/cli/test/qa-names-and-tags-run.test.ts",
    "packages/cli/test/qa-tags-selection.test.ts",
    "packages/core/src/qa/runner.test.ts",
    "packages/core/src/qa/runner.ts",
    "packages/spec/liveness/README.md",
    "packages/spec/liveness/qa.json",
    "packages/spec/liveness/state-counts.md (generated by gen:liveness-counts)",
    "packages/spec/src/qa/testing.zod.ts (describe text only)",
    "content/docs/deployment/cli.mdx (round 1, text)",
    "docs/qa/platform-checklist/areas/cli.json (round 1, text; revision 4 to 5)",
    "packages/spec/scripts/liveness/check-liveness.mts (round 1, comment only)"
  ],
  "deviations": [
    "The order named the knownGap at cli.json:456. The same item's source note, 'packages/spec/liveness/qa.json#tags (… its dead tags/requires rows are why no scenario selection exists)', was false for the same reason, so it was corrected in the same edit. It is in the same file and item, so it stays inside the amended surface.",
    "The '## Rework round 1' section sits under the 'Part of #20289' and 'Clause-②: no' lines rather than above them, so the closing-keyword guard and Check Changeset keep reading the same opening lines. The same single issue_patch also replaced the round-0 Acceptance-notes bullet ('outside the claimed file surface, not edited'), which this round made false.",
    "The first gate pass overlapped the prerequisite rebuild, because the unlocked gates ran beside a locked build. Every family that could read dist was therefore re-run in a clean second pass. Only check:pm-dispatch-gates, which reads no built output, keeps its first-pass reading.",
    "Carried from round 0: the PR opens with 'Part of #20289', not 'Fixes'; core is minor rather than patch because TestResult gains 3 public fields, and all three packages share one changesets fixed group; the skip-with-reason pin and requires wait on the decision; origin/main is not merged in, and main's newer commits do not touch this diff's files.",
    "The verify lock was held long: round 1 held it for 21m05s, 68 turbo tasks rebuilt from scratch because the worktree was re-created. Round 0 sessions held it for 16m35s, 16m57s and 13m37s.",
    "Cleanup: node_modules was removed before this report posted. The worktree (git worktree remove, never --force) is removed right after the comment lands, because post-stamped runs from it."
  ],
  "mcp_calls": "0",
  "api_writes": "Round 1: 2 REST writes through the fleet-write relay (POST /repos/objectstack-ai/objectstack/dispatches, 204, then the relay run). (1) issue_patch, PATCH /repos/objectstack-ai/objectstack/issues/20341 body, run 36367615540, read back byte-identical at 13549 bytes. (2) This os-dev-report comment, POST /repos/objectstack-ai/objectstack/issues/20289/comments via post-stamped. There was also 1 git push, 5d31b4d587, which is not a REST write. Round 0 had 3 relay writes: pr_create run 36360895587, the assignees label-write run 36360935834, and the os-dev-report comment 5861105453. It also had 6 git pushes.",
  "open_questions": [
    {
      "question": "How should TestScenario.requires (params, plugins) be honoured? What does 'met' mean for each sub-key, or should the block be retired? The family verdict is ENFORCE, but neither sub-key has an honest judge from os test today. plugins: the runner reaches the target only over HTTP. No server surface lists loaded plugins: DiscoverySchema advertises services and capabilities, and /.well-known/objectstack advertises neither, while kernel.getPlugins() is in-process only. The plugin spelling (npm name or plugin.name) is also undefined. params: 'environment variables or parameters' names no process. The runner's env is observable but unconsumable, because interpolation reads only captured context. The server's env is unobservable. Measured producers: zero. The repo's one suite, examples/app-showcase/qa/platform-smoke.test.json, declares no requires. Its only precondition is auth, written as description prose: 'Requires --token'.",
      "options": [
        "A: Enforce plugins against a NEW discovery field listing loaded plugins, which needs a DiscoverySchema field, a runtime producer from the kernel, and REST/dispatcher parity. Enforce params as env vars set in the process running os test, the JUnit @EnabledIfEnvironmentVariable / pytest skipif reading. An unmet requirement becomes SKIPPED with its reason, counted separately, never passed, and a failure-free run still exits 0. Costs: a new public discovery field; a disclosure call, because discovery is served unauthenticated on many hosts and would publish the stack's plugin inventory; a plugin-name spelling ruling; and a SKIPPED status in core and the CLI. Four axes: (1) business need: speculative, with zero producers, and the one real precondition is auth, which A does not cover. (2) Long-term: adds a public surface. (3) Anti-AI-error: plugin spellings invite guesses, so it needs a closed vocabulary to be safe. (4) Startup focus: the widest expansion.",
        "B: Replace plugins with services, judged against the discovery services map the runner already probes once per run (enabled true and status 'available'). Keep params as runner env, the same reading as A. Retire plugins immediately with a retiredKey prescription pointing to services: zero producers were measured, so there is no staged window. An unmet requirement becomes SKIPPED with its reason, naming what the target advertises. Costs: a spec narrowing (Clause-② yes (narrowing), a BREAKING banner and an ADR-0087 disposition) and the same SKIPPED status. There is no new server surface. Four axes: (1) business need: answers the question authors actually ask, 'is the AI/search surface served here?', from a contract that already exists. (2) Long-term: reuses ADR-0076 D12, 'advertise only what is mounted'. (3) Anti-AI-error: the vocabulary is the server's own service keys, and a misspelling skips with the advertised keys named instead of passing. (4) Startup focus: one reader and no new surface.",
        "C: Retire requires whole: a retiredKey tombstone on TestScenarioSchema, with the prescription 'select with --tags; state preconditions in description'. The ledger rows become tombstones. Costs: the smallest diff, but it drops a capability the mainstream has (JUnit @EnabledIfEnvironmentVariable, pytest skipif, Odoo tests scoped to installed modules). That contradicts the family's ENFORCE verdict for this key under the maintainer's criterion, which asks whether the mainstream has it, not whether anyone in the repo reads it. Four axes: (1) business need: nothing is lost today. (2) Long-term: no half-protocol. (3) Anti-AI-error: the strictest option, since the tombstone refuses the key loudly. (4) Startup focus: the best fit for a zero-pull declared surface."
      ],
      "recommendation": "B. It keeps the maintainer's criterion: the mainstream has declared preconditions, so build the consumer once. Every judgement it makes is against something the runner can observe today: discovery services, which it already fetches, and its own process env. It adds no server surface, which is A's cost. It removes the one sub-key that cannot be judged honestly rather than keeping it. The trade-off, stated plainly: C wins the anti-AI-error and startup-focus axes outright. If the maintainer weighs zero measured producers above the mainstream criterion, C is the call. Either way, the SKIPPED-with-reason leg lands only after this ruling."
    }
  ],
  "out_of_scope_findings": [
    "none new. The three round-0 carrier items (cli.mdx, the platform-checklist knownGap and the check-liveness.mts comment) are corrected in this round, after the seat amended the surface in 5861125164."
  ]
}

---

Generated by Claude Code

[objectstack-fleet]

os-dev-report

{
  "issue": 20289,
  "status": "needs_decision",
  "branch": "claude/issue-20289-os-test-names-tags-requires",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/20341",
  "session": "session_01QcAS3qiYYZNezaxZxaUdMV (mode:subagent, the PM's id, stamped on every commit's Claude-Session trailer; identity = the branch)",
  "premise_still_valid": true,
  "summary": "The landing round after contract review 5862007793 (PASS at 5d31b4d587) is done. The new head is 5b15c28fb1. There is no behaviour change and no requires work. (1) origin/main at d498113b50 is merged in with scripts/pm/os-regen-merge.sh as merge commit 9056455134, with no rebase, amend or force-push. The one conflict was packages/spec/liveness/README.md, and it keeps main's translation and rest_api rows plus this PR's qa row. Each row was verified byte-equal to its source side. The script's step 2 took main's side of the generated state-counts.md. It was regenerated with gen:liveness-counts in 76e9a48821, never hand-resolved, and gives qa 8/1 on top of main's totals, 940/5/1/162/9/1117. That commit discharged the merge's regen deferral. Step 4 ran on a fresh spec build: 'All 15 generated artifacts are up to date', so nothing else needed regenerating. git diff d498113b50 HEAD touches only this PR's 14 files, and main's lines in the three shared files survive with 0 removals in cli.mdx. (2) Commit 5b15c28fb1 carries the reviewer's nits, text only. (a) cli.mdx, the checklist knownGap and the changeset now say the unmet requirement 'surfaces only as whatever failure it causes, if any'; the scenario can pass. The liveness README qa row carried the same inaccuracy ('fails as an unexplained HTTP error') and got the same fix. (b) The checklist revision-5 history notes the id is bare when the name equals it. (c) The README qa row uses qa.json's wording: suiteName 'on every TestResult the suite produces'. (3) One PR-body issue_patch, because the merge falsified one sentence: 'Branch base: the branch sits on 10ea9eb2ed … The merge ref is left to CI.' now names the merge. It read back byte-identical at 13777 bytes. mergeable_state went from dirty to unstable, with CI still running. requires stays with the maintainer, and open_questions is unchanged.",
  "tests": "Landing round, at HEAD 5b15c28fb1, on the merged tree. check:liveness is green with 'qa 9 classified (live 8, dead 1)' and 'state-counts.md is current'. check:platform-checklist: 'OK — 15 areas, 266 items'. spec check:generated after a fresh build: 'All 15 generated artifacts are up to date', exit 0. main also moved packages/cli and packages/spec, so the qa pins re-ran on the merged tree: core src/qa/runner.test.ts 20 of 20; cli unit qa-tags-selection plus qa-suite-schema-load 19 of 19; cli integration qa-names-and-tags-run 8 of 8. This round changed no code. Package-wide test and typecheck on the merge ref belong to CI. Earlier rounds are unchanged: round 1's check-liveness own tests 262 of 262; round 0's core 1483 tests, the cli unit tier, typechecks and ablation, as reported in 5861105453 and 5861913911.",
  "gates": "Landing round, at 5b15c28fb1. dispatch-gates --repo objectstack-ai/objectstack --commands for the merged diff: 14 paths against merge base d498113b5, the same 118 families as round 1 with none added or dropped. The recreated worktree had no dist, so the prerequisites were rebuilt under the lock with the same 13 filters: exit 0, lock held 18m51s, a shared-box figure. All 118 then ran with nothing building in the tree, and all 118 exited 0. That includes the doc gates (doc-anchors, docs-*, doc-frontmatter, doc-route-spelling, docs-section-name, doc-authoring, affected-docs, drift-comment, cli-examples-parity), check:platform-checklist, check:liveness, check:generated, api-surface, dts-closure, dual-build-cjs-loads, the i18n trio and type-check-debt. check:pm-dispatch-gates took 1356s. --ran reconciliation, with every line 'COMMAND :: exit CODE', gave '118 derived, 118 run, 0 NOT-MEASURED, 0 UNRUN' (a derived zero). CI, read once after the push: 32 check-runs on 5b15c28fb1, 19 success, 2 skipped, 11 in_progress, 0 failed. That is in_progress, and the convergence wait belongs to the PM.",
  "line_budget": "n/a: no skills/** and no governed surface in the diff",
  "files_changed": [
    ".changeset/20289-os-test-names-tags.md",
    "content/docs/references/qa/testing.mdx (generated by check:generated --fix)",
    "packages/cli/src/commands/test.ts",
    "packages/cli/test/qa-names-and-tags-run.test.ts",
    "packages/cli/test/qa-tags-selection.test.ts",
    "packages/core/src/qa/runner.test.ts",
    "packages/core/src/qa/runner.ts",
    "packages/spec/liveness/README.md",
    "packages/spec/liveness/qa.json",
    "packages/spec/liveness/state-counts.md (generated by gen:liveness-counts)",
    "packages/spec/src/qa/testing.zod.ts (describe text only)",
    "content/docs/deployment/cli.mdx (round 1, text)",
    "docs/qa/platform-checklist/areas/cli.json (round 1, text; revision 4 to 5)",
    "packages/spec/scripts/liveness/check-liveness.mts (round 1, comment only)"
  ],
  "deviations": [
    "The merge commit 9056455134 carries git's default message, 'Merge remote-tracking branch origin/main into claude/issue-20289-os-test-names-tags-requires', and no Claude-Session / Co-Authored-By trailer pair. I committed the resolution with --no-edit, as the regen script prescribes. The order forbids amend, so it stands. The pre-push trailer check passed: no model identifier appears anywhere. The two commits after it carry the model-free pair.",
    "One nit reached a fourth file. The reviewer listed cli.mdx, cli.json and the changeset for the 'fails on whatever the missing plugin causes' inaccuracy. The liveness README qa row, which this PR rewrote and which nit (c) already touched, said the same thing as 'fails as an unexplained HTTP error', so it got the same correction.",
    "The PR body was patched once. The order allows that only if the merge made a sentence false, and the 'Branch base … sits on 10ea9eb2ed … merge ref is left to CI' bullet was false after the merge. Only that bullet changed.",
    "Following AGENTS' push-before-minutes rule, I pushed the three commits (9056455134, 76e9a48821, 5b15c28fb1) once, before the 18-minute rebuild. Their gates and step-4 check:generated then ran at that same head, so no second push was needed.",
    "Carried from earlier rounds: the PR opens with 'Part of #20289'; core is minor; the skip-with-reason pin and requires wait on the maintainer's decision.",
    "Cleanup: node_modules was removed before this report posted. The worktree (git worktree remove, never --force) is removed right after the comment lands, because post-stamped runs from it."
  ],
  "mcp_calls": "0",
  "api_writes": "Landing round: 2 REST writes through the fleet-write relay (POST /repos/objectstack-ai/objectstack/dispatches, 204, then the relay run). (1) issue_patch, PATCH /repos/objectstack-ai/objectstack/issues/20341 body, correcting one bullet, run 36373561351, read back byte-identical at 13777 bytes. (2) This os-dev-report comment, POST /repos/objectstack-ai/objectstack/issues/20289/comments via post-stamped. There was also 1 git push of 3 commits, head 5b15c28fb1, which is not a REST write. Earlier rounds: round 0 had 3 relay writes and 6 pushes; round 1 had 2 relay writes (issue_patch run 36367615540, comment 5861913911) and 1 push.",
  "open_questions": [
    {
      "question": "How should TestScenario.requires (params, plugins) be honoured? What does 'met' mean for each sub-key, or should the block be retired? The family verdict is ENFORCE, but neither sub-key has an honest judge from os test today. plugins: the runner reaches the target only over HTTP. No server surface lists loaded plugins: DiscoverySchema advertises services and capabilities, and /.well-known/objectstack advertises neither, while kernel.getPlugins() is in-process only. The plugin spelling (npm name or plugin.name) is also undefined. params: 'environment variables or parameters' names no process. The runner's env is observable but unconsumable, because interpolation reads only captured context. The server's env is unobservable. Measured producers: zero. The repo's one suite, examples/app-showcase/qa/platform-smoke.test.json, declares no requires. Its only precondition is auth, written as description prose: 'Requires --token'.",
      "options": [
        "A: Enforce plugins against a NEW discovery field listing loaded plugins, which needs a DiscoverySchema field, a runtime producer from the kernel, and REST/dispatcher parity. Enforce params as env vars set in the process running os test, the JUnit @EnabledIfEnvironmentVariable / pytest skipif reading. An unmet requirement becomes SKIPPED with its reason, counted separately, never passed, and a failure-free run still exits 0. Costs: a new public discovery field; a disclosure call, because discovery is served unauthenticated on many hosts and would publish the stack's plugin inventory; a plugin-name spelling ruling; and a SKIPPED status in core and the CLI. Four axes: (1) business need: speculative, with zero producers, and the one real precondition is auth, which A does not cover. (2) Long-term: adds a public surface. (3) Anti-AI-error: plugin spellings invite guesses, so it needs a closed vocabulary to be safe. (4) Startup focus: the widest expansion.",
        "B: Replace plugins with services, judged against the discovery services map the runner already probes once per run (enabled true and status 'available'). Keep params as runner env, the same reading as A. Retire plugins immediately with a retiredKey prescription pointing to services: zero producers were measured, so there is no staged window. An unmet requirement becomes SKIPPED with its reason, naming what the target advertises. Costs: a spec narrowing (Clause-② yes (narrowing), a BREAKING banner and an ADR-0087 disposition) and the same SKIPPED status. There is no new server surface. Four axes: (1) business need: answers the question authors actually ask, 'is the AI/search surface served here?', from a contract that already exists. (2) Long-term: reuses ADR-0076 D12, 'advertise only what is mounted'. (3) Anti-AI-error: the vocabulary is the server's own service keys, and a misspelling skips with the advertised keys named instead of passing. (4) Startup focus: one reader and no new surface.",
        "C: Retire requires whole: a retiredKey tombstone on TestScenarioSchema, with the prescription 'select with --tags; state preconditions in description'. The ledger rows become tombstones. Costs: the smallest diff, but it drops a capability the mainstream has (JUnit @EnabledIfEnvironmentVariable, pytest skipif, Odoo tests scoped to installed modules). That contradicts the family's ENFORCE verdict for this key under the maintainer's criterion, which asks whether the mainstream has it, not whether anyone in the repo reads it. Four axes: (1) business need: nothing is lost today. (2) Long-term: no half-protocol. (3) Anti-AI-error: the strictest option, since the tombstone refuses the key loudly. (4) Startup focus: the best fit for a zero-pull declared surface."
      ],
      "recommendation": "B. It keeps the maintainer's criterion: the mainstream has declared preconditions, so build the consumer once. Every judgement it makes is against something the runner can observe today: discovery services, which it already fetches, and its own process env. It adds no server surface, which is A's cost. It removes the one sub-key that cannot be judged honestly rather than keeping it. The trade-off, stated plainly: C wins the anti-AI-error and startup-focus axes outright. If the maintainer weighs zero measured producers above the mainstream criterion, C is the call. Either way, the SKIPPED-with-reason leg lands only after this ruling."
    }
  ],
  "out_of_scope_findings": [
    "none new."
  ]
}

---

Generated by Claude Code

[objectstack-fleet]

ACCEPT (four of five keys) — PR #20341 at head 5b15c28fb1 · domain:spec seat 2 (session_01QcAS3qiYYZNezaxZxaUdMV) · 2026-09-28T03:30Z

Seat review against GitHub, not the report:

• Shape: Part of #20289 (the card stays open for requires), a standalone Clause-②: no line, 14 files, +624/−42. NOT governed.
• Delivered: triage 5859588706 ENFORCE, for four of the family's five keys.
  • QA.TestResult carries scenarioName / description (on both envelopes) and suiteName (on every result a suite produces). os test prints NAME (FILE) per suite, NAME [ID] per scenario, and a failed scenario's description.
  • --tags selects scenarios: any-of, exact, case-sensitive. Deselected scenarios are counted and never passed. An unknown tag is named. An empty entry is refused at parse. No-match exits 0, or 1 under --fail-on-empty.
  • The ledger rows qa.name, scenarios.name, scenarios.description and scenarios.tags are live, citing their readers.
  • requires' describe says NOT CHECKED, and its row stays dead.
  • Ablation: 3 of 13 unit and 3 of 8 integration went red, with the controls green, then restored.
• REWORK round 1 (5861125164): corrected the three texts this PR falsified (the cli.json knownGap, cli.mdx os test, the check-liveness.mts header).
• The landing round: merged main and regenerated state-counts.md, and carried the reviewer's accuracy nits.
• File-surface amendments to claim 5860293954 (recorded here, ⛔ not a second claim): those three texts, plus the liveness/README.md qa row.
• At-tier contract review:
  • PASS 68/68 5862007793 at 5d31b4d587;
  • delta PASS 94/94 at this head, posted on the PR in this window.
  • Nothing reads a filtered-out or erroring scenario as passed. No text claims requires does anything.
• CI: 38 success, 4 roster skips, 0 failures. check-expected-skips is OK.
• Landing: mergeable_state: clean, and the driver-free merge-tree against origin/main is clean. Flipped to ready with auto-merge in this window.
• Next on this card: once this lands, the card moves pm:dispatched → needs-user-decision for requires, with the four-axis analysis and the options A / B / C from the dev's reports (5861105453, 5861913911).