os-dev contract: three test obligations narrowed — ablation is a one-time proof, import-side tests only on public-surface change, no wording pins (maintainer-directed, part C)

[claude]

Filed by the skills lane seat (session session_019RfFHiRCSs3JXLK4cwcfox, os-steve) on the maintainer's direction, 2026-09-07T02:1xZ. domain:skills; governed (.claude/agents/os-dev.md and any reference the obligations live in).

Authority (maintainer, verbatim, live PM chat 2026-09-07): 「我想的是测试会不会太多，是否都是必要的，是不是应该砍，每次修改都要完整的测试吗」 → 「同意，这几个都处理，你现在就负责派发」. This card is part C.

What

Three test obligations in the dev contract are narrowed, because they are the source of the growth curve (35–89 new test files a day; 3,388 test files against 2,183 source files):

1. Reverse verification and ablation are one-time proofs, quoted in the PR body and the report — never a permanent test file. The legs stay mandatory; what changes is that their evidence is the quoted before/after run, not a fixture that stays in the tree.
2. Import-side package tests are owed only when a PUBLIC surface changes — a packages/spec contract, a published exports entry, a wire shape. A change that keeps every public surface byte-identical owes its own package's tests and the derived gates, not a test in every package that imports the module.
3. No pin tests on message wording. A remedy string, a verdict line or an error text is pinned only where the words are the operative criterion a consumer parses; otherwise the assertion is on the kind, the exit code or the named subject, never on the prose.

Measured

• os-dev.md line 130 (「不按改了哪些包猜:普通 import 被改模块的包也在清单里,欠它们的测试」) is the import-side obligation; lines 188–259 are the reverse-verification and ablation block; the wording-pin habit has no single line in os-dev.md and may live in references/review-checklist.md or core-rules.md — the flight measures where each obligation is stated before folding, and every carrier moves in the same PR.
• The line pins hold: os-dev.md 403/403, core-rules.md 150/150; every change is a fold under the pins, rules-only, plain language, one rule per line (the maintainer's 2026-09-04 rulings).

Ruling

• Governed: draft PR, in-seat contract-tier review, os-zhuang + hotlong, needs-user-decision, the final 维护者速读, human merge.
• Serial: same file as os-dev: the "build spec" verification step passes without measuring anything off packages/spec/src — condition it on the file surface or replace it by the dispatch-gates derivation (#16104 section 3.1) #16388 (which waits behind this card now) and as PR agents(os-dev): the verify-lock waiting discipline as protocol, and the finding admission threshold #16375 (merged 02:02:55Z — branch from origin/main at or after 6eba38f5a3).
• Clause-②: no — the contract narrows what a dev owes; no published surface moves.

Acceptance

• check-skill-line-ratchet live and self-test green at unchanged ceilings; check-governed-merges --test exit 3; the corpus gates (check:skill-frame-sync, check-clause2-carriers, check-skill-id-lint) green.
• The PR body carries, per obligation, the old line, the new line, and the sentence that says what a dev no longer owes.

---

Generated by Claude Code

[claude]

Claim: PM loop round 9 — flight Y, maintainer-directed (「同意，这几个都处理，你现在就负责派发」, 2026-09-07): three test obligations in the dev contract narrowed, rules-only, under the unchanged line pins — reverse verification and ablation are one-time proofs quoted in the PR and the report, never a permanent test file; import-side package tests are owed only when a public surface changes; no pin tests on message wording unless the words are a consumer's parsing criterion. Every carrier of each obligation is located by measurement before folding and moves in the same PR.
Session: session_019RfFHiRCSs3JXLK4cwcfox
Branch: claude/issue-16456-dev-contract-test-obligations
Worktree: objectstack-issue-16456
Domain: domain:skills
File surface: .claude/agents/os-dev.md (403/403) and whichever of .claude/skills/pm-dispatch/references/core-rules.md (150/150), references/review-checklist.md or SKILL.md (811/811) carries one of the three obligations — measured, not assumed. ⛔ No other line moves; folds only.
Container & model: S, mode:subagent, fable — dispatch-gates.mjs --tier: MANDATORY on .claude/agents/os-dev.md; on a fable 429 the seat's exemption applies, opus, recorded here.
Clause-②: no
Serial: PR #16375 (M) merged 02:02:55Z — branch from origin/main at or after 6eba38f5a3; no open PR touches os-dev.md; #16388 waits behind this card. Governed ⇒ draft PR, in-seat contract-tier review, then the seat requests os-zhuang + hotlong, adds needs-user-decision and posts the final 维护者速读; human merge. The PR carries Fixes #16456.

---

Generated by Claude Code

[claude]

os-dev-report

{
  "issue": 16456,
  "status": "done",
  "branch": "claude/issue-16456-dev-contract-test-obligations",
  "pr": "https://github.com/objectstack-ai/objectstack/pull/16460",
  "premise_still_valid": true,
  "summary": "One file, .claude/agents/os-dev.md, 403/403 with every line at or under 120 bytes, rules-only: ② L129–130 rewritten in place (affected = the changed package; importers owe tests only when a public surface — spec contract, published exports, wire shape — changes; a byte-identical public surface owes own-package tests and the derived gates only); ① the L210–214 commit-first block folds 5→4 and now opens with 'reverse verification and ablation are one-time proofs quoted in the PR body and the report, never a permanent test file'; ③ added as new L227 beside L228 — the measurement found no prior carrier of the wording-pin obligation anywhere in the corpus (it was a habit). Carriers outside the five named files (lanes/services.md L17, lanes/spec.md L38–39, dispatch-runbook.md L223–225, AGENTS.md L364–371, published dev-template.md L71 and L95) already read true under the ruling and are listed with file and line in the PR body; DoD test clause and both cli tier sentences read true unchanged. Deviations named rather than chosen silently: the commit carries the harness-written Co-Authored-By trailer (the AGENTS.md / dispatch-runbook exemption) while the PR title, body and this report carry no model identifier; the 速读 draft uses os-dev.md's five fixed headings with 席位意见 left blank (the brief listed four); the TURBO_SCM_BASE turbo ls --affected spelling left together with the CI-set definition it served (that definition is the echo the ruling replaces) and the L185 consumer-sweep filter remains the enumerator when a public surface does change. Draft PR #16460 opened, skip-changeset added through the additive endpoint with a compare read-back, no reviewer requested, no ready flip, no queue action; assignee was already set by the dispatch, the newest Claim: names this branch.",
  "tests": "Union run at ddb4213508 (the only and final commit). Derived: node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack, no paths (1 path vs merge base 6eba38f5a) → 16 families, every one run with the exit captured before any pipe, reconciled: '✓ dispatch-gates --ran: 16 derived famil(ies) accounted for — 16 run, 0 NOT-MEASURED.' (exit 0). pnpm check:pm-skill-ratchet exit 0 — '✓ check-skill-line-ratchet: .claude/agents/os-dev.md is 403 lines (ceiling 403; headroom 0).' and '✓ check-skill-line-ratchet self-test: 155 cases pass.'; LC_ALL=C awk over the file: max=120 over120=0. pnpm check:pm-skill-id-lint exit 0 — '✓ check-skill-id-lint: 26 file(s) clean'. pnpm check:skill-frame-sync exit 0 — '4 axes … binding sentence present in all 2; 4 count mention(s) agree'. node scripts/pm/check-clause2-carriers.mjs --self-test exit 0 — '190 cases pass'. node scripts/pm/check-governed-merges.mjs --test .claude/agents/os-dev.md exit 3 — 'One hit governs the whole PR … .claude/** ×1'. pnpm check:pm-governed-merges exit 0 (self-test 274 assertions). pnpm check:nul-bytes exit 0 — 'scanned 8078 text file(s) … no raw ASCII control bytes'. pnpm check:agent-model-declared exit 0. agent-test-spelling, doc-authoring, driver-memory-census, refd-timer-probe, watch-hint-literal, closing-keyword-parity (+ self-test), comment-mask-corpus, governed-queue-guard --self-test: all exit 0. check:doc-formula-expressions first exit 3 'PREREQUISITE NOT MET — the workspace package @objectstack/formula is not built' (nothing measured) → formula + lint built under the verify-lock (waited 384s for another holder, command-exit 0) → re-run exit 0 — '✓ check:doc-formula-expressions: 22 record-scoped formula example(s) across 430 files / 1371 TS blocks judged clean'. Whole-repo pnpm lint through bash scripts/pm/os-verify-lock.sh, slot issue-16456, one acquisition: 'os-verify-lock: VERDICT command-exit 0 · held the lock 89s (1m29s) · waited 0s' (shared-box seconds). The fold was proved on disk by grep -c of the six removed anchors (0 each) and six inserted anchors (1 each), never by the editor's exit code. git fetch + git merge-tree --write-tree HEAD origin/main at 245c6a23cd: exit 0, 0 CONFLICT lines. No ablation applies — the diff is instruction text. CI on head ddb4213508, polled in the foreground until every run completed: 29 check runs — 17 success, 12 skipped, 0 failure; Lint & Repo Gates success (completed 2026-09-07T03:05:28Z), TypeScript Type Check success, Test Core success, Dogfood Regression Gate success, Governed Surface Queue Guard success, Check Changeset success. Labels on PR 16460 at close (read back after CI): documentation · size/s · skip-changeset.",
  "mcp_calls": "1 — create_pull_request. The card and its comments came from the public issue page's embedded JSON payload; labels (additive POST + compare read-back), the PR read-back, the check-run poll and this report comment went through REST after a repo-scoped read probe answered 200 (core 15000/15000).",
  "open_questions": [],
  "out_of_scope_findings": []
}

---

Generated by Claude Code