plan-feature: lean fast-lane build path + mechanical proof-of-done (BO-2400, BO-2500)#378
Merged
Conversation
…400, BO-2500)
Product Owner stage of the fast-lane feature. Two L0 goals: BO-2400 (lean fast lane for scoped interactive work) with L1s BO-2400a-d (lean loop, path selection, caching efficiency, telemetry); BO-2500 (mechanical proof-of-done) with L1s BO-2500a-d (test-backed requirements, unskippable CI gate, real-artifact fixtures, retire opinion-only review). BO-2400 depends_on BO-2500 (the fast lane is only safe because mechanical proof replaces LLM gates). origin_agent BrainCandy, readiness draft. SKIP=check-product-truth-{validate,generate}: pre-existing store-wide index drift unrelated to these build-orchestration ACs (no product_truth blocks).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
… BO-2400/BO-2500)
Business Analyst stage: decomposed the 8 L1s into 39 testable L2/L3 ACs. Fast lane (BO-2400a-d): two-agent loop with agent-count==2 invariant, script-gated AC selection/red-baseline/green+coverage, path-selection rule, stable-prefix caching, per-invocation telemetry. Mechanical proof (BO-2500a-d): no done without a present+passing # covers:<id> test, unskippable CI gate on fresh checkout, real-artifact/round-trip fixtures, gate-agent retirement guarded by replacement existence. Each L2 depends_on its L1; L1 covered_by updated. origin_agent BrainCandy, readiness draft. SKIP=check-product-truth-{validate,generate}: pre-existing store-wide index drift, unrelated (no product_truth blocks).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…t-lane (BO-2400/BO-2500)
IT Product Owner stage: enriched all L2/L3 leaves with assigned_agent, it_requirements, estimated_complexity, delivers_to/expects_from contracts, doc_links (verified-existing architecture docs + ADRs), and test contracts; authored 13 documentation ACs from the L1 documentation_triggers. Assignments: python-coder for deterministic gates/scripts (mark_ac_done, test_enforcement, commit_guardian), llm-expert for workflow/CI/template surface, documentation-expert/architecture-diagram-author for docs. Criteria text unchanged. All readiness: draft (PR is the review vehicle). BO-2400a carries child_limit_override: 9 (5 behavior L2s + 3 doc ACs) pending a tree-split review. Caveats recorded in-AC: BO-2500b-3 required CI check is a GitHub ruleset admin action (test_required:false); BO-2500c-2/c-3 are test-writer template conventions. SKIP=check-product-truth-{validate,generate}: pre-existing store-wide index drift, unrelated (no product_truth blocks).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…proof TDD red baseline for the done-proof coverage engine (BO-2500a-1/a-2/a-3/a-1-i/a-2-i): 13 unittest cases in unit_tests/ac_store/test_bo2500a_done_proof.py defining the contract verify_done_eligible(ac_id, *, ac_root, test_root) + a mark_ac_done coverage gate. Fixtures use the real serializer (yaml.safe_dump) and real on-disk .py test files with genuine pass/fail/xfail/skip bodies (dogfoods BO-2500c fixture-authenticity). RED against current code (ImportError: verify_done_eligible not defined). Coder makes green next. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…2500a, green) Implements the done-proof gate (lean two-agent build, red baseline fba02ce → green). New scripts/ac_store/done_proof.py: verify_done_eligible(ac_id, *, ac_root, test_root) returns {eligible, reason, passing_tests, failing_tests, dangling_tags} — eligible iff >=1 test tagged '# covers:<id>' exists AND all such tests PASS (xfail/skip/xpass/error count as non-passing; determined by running real pytest and parsing -v output, not exit code). Reuses extract_covers_tag. mark_ac_done() gains test_root= gate: refuses (exit 3) with a reason when ineligible; back-compat when test_root is None (call-site audit clean). All 13 BO-2500a tests green; ac_store suite 396 passed / 0 new failures. Closes BO-2500a-1/a-2/a-3/a-1-i/a-2-i (work_status done, implemented_by set). SKIP=check-product-truth-{validate,generate}: pre-existing unrelated index drift. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…2500c-1/c-2)
Adds section 2h.2 'Fixture Authenticity Rule' to templates/agents/test-writer.md: fixtures for serialized formats (YAML/JSON/etc.) must be produced by the real serializer or read from a real on-disk artifact — never a hand-typed literal; parser/validator tests must round-trip through the real producer. Rationale cites the EPIC-PhantomDoneFilesTouched column-0-vs-indented no-op. Closes BO-2500c-1/c-2 (work_status done; test_required:false prompt-convention ACs). Dogfooded this session: the BO-2500a red tests were authored under this rule (yaml.safe_dump AC fixtures + real .py covers-test files, no mocked pass/fail). SKIP=check-product-truth-{validate,generate}: pre-existing unrelated index drift.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…t bundle
Adds two refinements surfaced while dogfooding the lean build: (1) BO-2400e (L1) + BO-2400e-1/e-2 — the lean lane retains git isolation (branch off latest origin/main, PR-only) and incremental logical commits (red baseline -> green -> each change), dropping orchestration overhead, not git discipline. (2) BO-2400c-5 (L2) — build agents receive a layered context bundle (high-level L0/L1 + batch ACs + prior area tests + architecture) ordered stable-first by change-frequency for cacheability. Parent covered_by updated (BO-2400 += BO-2400e; BO-2400c += BO-2400c-5). readiness: draft. SKIP=check-product-truth-{validate,generate}: pre-existing unrelated index drift.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…cement surfaces TDD red baseline for BO-2500b-1/b-1-i/b-2: 25 tests across unit_tests/commit_guardian/test_bo2500b_done_proof_hook.py (pre-commit static covers-tag check + CI-authoritative verify_done_eligible check + mark_ac_done --test-root CLI) and unit_tests/build/test_bo2500b_ci_workflow.py (ci.yml done-proof job). Real fixtures: yaml.safe_dump AC files, real .py covers-test files, and the real on-disk commit_guardian.json/ci.yml parsed (not hand-typed) per BO-2500c. RED in isolation (check_done_proof absent; no ci.yml job). Coder makes green next. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…rk_ac_done --test-root (BO-2500b, green) Green for BO-2500b-1/b-1-i/b-2 (red baseline 9aeeba9). New templates/scripts/commit_guardian/check_done_proof.py: check_staged_done_proofs (fast, static, diff-aware covers-tag presence on staged newly-done ACs — skippable pre-commit) and check_all_done_acs (CI-authoritative via verify_done_eligible, runs real tests). Registered in commit_guardian.json (id check-done-proof, no always_run — skippable); auto-deployed via build.py's templates/scripts/commit_guardian scan; repo root resolved via git rev-parse to avoid the deployed-layout no-op trap. .github/workflows/ci.yml gains a done-proof job (checkout -> build -> check --mode ci) so a skipped pre-commit is still caught on a fresh checkout (b-1-i). mark_ac_done gains --test-root wiring the coverage gate to the CLI. 30 tests green (23 hook + 7 ci); ac_store suite 111/111, no regressions. b-3 (required, non-bypassable merge check) is an ops/ruleset action, left open. SKIP=check-product-truth-{validate,generate}: pre-existing unrelated index drift. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
TDD red baseline for BO-2400b-1/b-2/b-3/b-3-i/b-3-ii: 24 tests in unit_tests/workflows/test_bo2400b_path_selection.py defining a pure choose_lane(*, scope, attended, defect_cost, override) -> {lane, reason, ambiguous, overridden} contract (scoped+interactive+low-defect -> fast; large|unattended|high-defect -> heavy; unknown scope -> heavy+ambiguous; explicit override wins). RED (ModuleNotFoundError: path_selection). Coder makes green next.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
New scripts/build_orchestration/path_selection.py: pure choose_lane(*, scope, attended, defect_cost, override) -> {lane, reason, ambiguous, overridden}. Deterministic single documented rule: scoped+interactive+low-defect -> fast; large|unattended|high-defect -> heavy; unknown scope -> heavy + ambiguous (reason recorded); explicit override wins (recorded). 26 tests green. Closes BO-2400b-1/b-2/b-3/b-3-i/b-3-ii. SKIP=check-product-truth-{validate,generate}: pre-existing unrelated index drift.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
TDD red baseline for BO-2400d-1/d-2/d-3/d-1-i: 22 tests in unit_tests/feedback/test_bo2400d_telemetry.py. Contract: scripts/agent-health/agent_telemetry.py::emit_agent_telemetry(record, *, sink_path) appends one JSONL record {ts, lane, agent, duration_ms, tokens_in, tokens_out, cache_read_tokens, unit_id}, WARNs + bumps a failed-write counter on OSError (never silent); get/reset_failed_write_count; and generate_health_report.py::build_lane_comparison_report(sink) for fast-vs-heavy per-unit aggregates. Real fixtures via the real emitter to a tmp sink + a genuine unwritable path. RED (ModuleNotFoundError: agent_telemetry). Coder makes green next.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…ort (BO-2400d, green)
New scripts/agent-health/agent_telemetry.py: emit_agent_telemetry appends one JSONL record {ts, lane, agent, duration_ms, tokens_in, tokens_out, cache_read_tokens, unit_id}; on OSError logs a WARNING + increments a failed-write counter (get/reset_failed_write_count) and does NOT raise — fail-loud-not-fatal per BO-2400d-1-i. generate_health_report.py gains build_lane_comparison_report(sink) -> per-lane fast-vs-heavy aggregates (count, durations, tokens, per-unit cost/time; graceful single-lane), and its pre-existing _load_jsonl open() is now wrapped per the exception-handling gate (IO-001). 26 tests green; feedback suite 61/61. Closes BO-2400d-1/d-2/d-3/d-1-i. Token/duration capture is harness-fed; this is the sink/emit/report infra. SKIP=check-product-truth-{validate,generate}: pre-existing unrelated index drift.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
TDD red baseline for BO-2400c-1/c-3/c-5 (the in-repo parts; c-2 1h-TTL is harness config, out of code scope): 18 tests in unit_tests/workflows/test_bo2400c_context_bundle.py defining assemble_context_bundle(*, architecture, conventions, high_level, acs, prior_tests, prior_outputs, working_diff, breakpoint_marker) in scripts/injection_builders.py — stable layers (architecture->conventions->high_level) before a single cache breakpoint, volatile layers (acs->prior_tests->prior_outputs->working_diff) after, ordered by change-frequency; asserts the stable prefix is byte-identical across invocations with differing volatile inputs (cacheable-prefix property) and prior outputs land in the variable suffix. RED (ImportError: assemble_context_bundle). Coder makes green next. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…-2400c, green)
scripts/injection_builders.py gains assemble_context_bundle(*, architecture, conventions, high_level, acs, prior_tests, prior_outputs, working_diff, breakpoint_marker): stable layers (architecture->conventions->high_level) before one cache breakpoint, volatile layers (acs->prior_tests->prior_outputs->working_diff) after, ordered by change-frequency. Stable prefix is byte-identical across calls with differing volatile inputs (cacheable-prefix property); prior-phase outputs threaded into the variable suffix. 18 tests green; prompt_assembly 36/36, no regressions. Closes BO-2400c-1/c-3/c-5. BO-2400c-2 (1h cache TTL) left open: harness/API config, not repo code. SKIP=check-product-truth-{validate,generate}: pre-existing unrelated index drift.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…unner TDD red baseline for BO-2400a-1..a-5 + L3s: 19 behavioral tests (unit_tests/build_orchestration/test_bo2400a_fast_lane.py) for a deterministic scripts/build_orchestration/fast_lane.py — select_batch (deterministic, capped), verify_red_baseline (all-red gate + offender), verify_green_and_coverage (green + covers gate + still-red report) reusing scan_ac_store + done_proof; and 17 structural tests (unit_tests/workflows/test_bo2400a_runner_structure.py) asserting a new templates/workflows-js/fast-lane-build.js dispatches EXACTLY two agents (test-writer + coder), references the deterministic gates in order, and contains NO ticket-supervisor / planner-agent / per-ticket-worktree constructs. Real fixtures (yaml.safe_dump ACs + real pytest); .js tests parse the real file. RED. Coders make green next. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
… green)
The centerpiece. scripts/build_orchestration/fast_lane.py: select_batch (deterministic ready-AC batch, capped; reuses scan_ac_store filter/sort), verify_red_baseline (all-red gate + offender), verify_green_and_coverage (green + covers gate + still-red report; reuses done_proof.verify_done_eligible). templates/workflows-js/fast-lane-build.js: lean E2 runner dispatching EXACTLY two agents (test-writer -> red-baseline gate -> coder -> green+coverage gate -> stage), no supervisor/planner/per-ticket-worktree. 19 behavioral + 17 structural tests green; fast_lane behaviorally spot-checked against the real AC store; ruff clean; node --check clean. Closes BO-2400a-1/a-2/a-3/a-4/a-5/a-2-i/a-3-i/a-4-i (a-6/7/8 are doc ACs, pending). SKIP=check-product-truth-{validate,generate}: pre-existing unrelated index drift.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…BO-2500d) BO-2500d is realized by construction — BO-2400a's fast-lane-build.js was built WITHOUT the opinion-only gate agents, so this slice adds 23 regression GUARD tests (unit_tests/workflows/test_bo2500d_gate_retirement.py) that lock the property in: fast lane excludes pr-reviewer/ac-validator/ac-fulfillment-gate/change-scope-reviewer (d-1); heavy build-feature.js/build-ticket.js retain their review gates (d-2); fast lane has the mechanical red-baseline + green+coverage gates as arbiters (d-3); and a guard that those mechanical gates cannot be removed (d-1-i). All 23 green (property holds; no red phase — honest). Parses the real .js files. Closes BO-2500d-1/d-2/d-3/d-1-i. NOTE: d-2 asserts pr-reviewer/change-scope-reviewer on this branch; ac-validator/ac-fulfillment-gate live on PR #375, unmerged here. SKIP=check-product-truth-{validate,generate}: pre-existing unrelated index drift. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…s (13 doc ACs)
Full documentation set for the fast lane, authored by documentation-expert specialists against the real code. How-tos (6): fast-lane-build, choose-build-path, compare-build-lanes, prove-ac-done, done-proof-enforcement, real-artifact-fixtures. Diagrams (4): c3 fast-lane build-loop + c2 fast-lane components + c3 done-proof sequence + c2 fast-vs-heavy phases (bidirectional parent/children links). References (3): fast-lane-prompt-caching, build-telemetry, fixture-policy. Closes all 13 doc ACs (BO-2400a-6/a-7/a-8, b-4, c-4, d-4/d-5; BO-2500a-4/a-5, b-4, c-4/c-5, d-4). Fixed invalid flight_level on two diagrams; removed two pre-existing dead related_docs links in agent_delivery_workflows.md. SKIP=check-doc-frontmatter: all feature docs pass — the sole residual violation is a PRE-EXISTING infra issue (the docs/INDEX.md generator emits no last_updated), unrelated to this feature; CI re-checks. SKIP=check-product-truth-{validate,generate}: pre-existing unrelated index drift.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
….md)
Brings the fast-lane branch current with origin/main. Only conflict was the auto-generated docs/INDEX.md, resolved by regenerating it via scripts/generate_doc_index.py so it reflects both branches' docs. build-orchestration.md and test-writer.md auto-merged cleanly. SKIP=check-doc-frontmatter: pre-existing INDEX.md generator omits last_updated (infra, unrelated); SKIP=check-product-truth-{validate,generate}: pre-existing index drift.
…fects + revert overclaimed done flags
Independent code-review + logic-check found the fast-lane RUNNER is phantom-done (engine is real). Red tests added: test_fast_lane_cli.py (H-2: fast_lane.py needs a real CLI — behavioral), test_bo2400a_runner_wiring.py (H-1: fast-lane-build.js must actually CONSUME the gate invocations, guard the coder dispatch on the red result, gate the ok-return on the green result, and wire choose_lane/emit_agent_telemetry/assemble_context_bundle — semantic, not grep), test_agent_telemetry_failsafe.py (M-1: json.dumps outside try lets TypeError escape), test_check_done_proof_ci_fail_closed.py (M-2: CI mode fails open), test_resolve_root_git_preferred.py (M-4: root via __file__ not git rev-parse). Reverted BO-2400a-1/a-2/a-3/a-4/a-5 and BO-2500b-1-i work_status done->todo (their done was overclaimed — runner gates never executed; CI check is continue-on-error so does not block). Coders make green next. SKIP=check-product-truth-{validate,generate}: pre-existing unrelated index drift.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…-1/H-2/M-1/M-2/M-4)
Makes the review's red tests green. H-2: fast_lane.py gains a real argparse CLI (select_batch/verify_red_baseline/verify_green_and_coverage → JSON + gate exit codes). H-1: fast-lane-build.js now actually executes its gates — the coder dispatch is guarded on the test-writer's red-baseline result, the final status:ok is guarded on the coder's green+coverage result, gates_passed is derived (not hardcoded), and choose_lane/emit_agent_telemetry/assemble_context_bundle are wired in (b/c/d no longer dormant). M-1: agent_telemetry json.dumps moved inside try + catches (OSError,TypeError) so a non-serializable record logs+counts instead of crashing the caller. M-2: check_done_proof CI mode fails CLOSED (exit non-zero) on checker error / missing 'eligible'. M-4: _resolve_root prefers git rev-parse --show-toplevel (templates source + deployed copy). Updated BO-2500d-3 test: assert the green gate RESULT is consumed after the coder (arbiter), not the invocation string's source position (the corrected wiring embeds it in the coder prompt). fast_lane CLI 12, runner wiring 8, telemetry 5, CI-fail-closed + git-root 8 — all green; workflows 92, build_orchestration 31, agent_health 5, ruff clean, no regressions. Re-flipped BO-2400a-1/a-2/a-3/a-4/a-5 to done (runner now real). BO-2500b-1-i stays todo (blocks-merge still needs the required-ruleset ops step). SKIP=check-product-truth-{validate,generate}: pre-existing unrelated index drift.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…vely (BP-1401)
next_seq() used a non-recursive Path.glob at arch_dir, but all C4 diagrams live under docs/architecture/diagrams/ — so it matched nothing and returned 1 for every level, causing sequence-number collisions. Changed to Path.rglob. The AC + regression test were authored earlier via /quick-fix and BP-1401 was marked done, but the actual code fix was never applied (test was RED) — a phantom-done surfaced during fast-lane review cleanup; the fix is now really in place and unit_tests/test_next_diagram_seq.py passes. SKIP=check-product-truth-{validate,generate}: pre-existing unrelated index drift.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…d-prep, INDEX generator, docs Closes the reviewer-flagged open items. (1) done-proof can be a REQUIRED gate: check_done_proof gains ci-changed mode (check_changed_done_acs — enforces only PR-changed done ACs, never legacy store debt), ci.yml done-proof job is now blocking (continue-on-error removed) + fetch-depth:0 + --mode ci-changed; 14 tests. Once added to the require-ci-lint ruleset (UI step), BO-2500b-1-i/b-3 are satisfied. (2) docs/INDEX.md now carries last_updated — the generator source was already fixed (3a2cb7b); build.py redeployed it + regenerated INDEX, so check-doc-frontmatter no longer needs a SKIP (this commit proves it — no check-doc-frontmatter skip). (3) agent-telemetry.md rewritten to the real agent_telemetry.py surface (stale emit_event.py removed, verified gone); new canonical docs/how-to/documentation/write-reference.md. SKIP=check-product-truth-{validate,generate}: pre-existing unrelated index drift. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
… now required)
Deploy-manifest gap: build_ac_store deployed a hardcoded list that omitted done_proof.py (created this session) — so the deployed check_done_proof hook + the now-REQUIRED CI done-proof gate crashed with ModuleNotFoundError: done_proof, which would have blocked every merge. Added done_proof.py to the build_ac_store deploy_map; build.py redeploys it. Caught live by the check_done_proof pre-commit hook firing on this very commit (dogfood). Also closes BO-2500b-1-i + BO-2500b-3: 'Proof-of-done coverage check (BO-2500b)' is now in the require-ci-lint ruleset (verified via gh api) — the gate blocks merge for any PR-changed done AC lacking a passing '# covers:' test. SKIP=check-product-truth-{validate,generate}: pre-existing unrelated index drift.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…F541) CI's required Lint (ruff) check failed on 8 violations in agent-written test files (unused imports, unused variable, extraneous f-prefixes) — the coders/test-writers didn't lint their new test files. Auto-fixed 7 via ruff --fix; removed one unused 'result' assignment in test_resolve_root_git_preferred.py manually (assertion checks mock_run.called, not the return value). All logic-neutral; ruff clean across scripts/unit_tests/tests. Unblocks the required Lint gate on PR #378. SKIP=check-product-truth-{validate,generate}: pre-existing unrelated index drift. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…, done_proof import, runner dual-engine) The required Test-suite (pytest) gate failed on #378 (42 real failures), caught by running the full suite (per-slice runs missed them). Three self-inflicted regressions from the review-remediation, fixed: (1) unit_tests/build/__init__.py made 'build' shadow scripts/build.py -> ~38 AttributeError cascade + shadow-guard failures; moved the two test files to unit_tests/build_guards/ and deleted the shadow. (2) check_done_proof import: fail-safe try/except (verify_done_eligible=None fallback) keeps a patchable module global + lazy None-fallback, so the module imports cleanly in the templates layout AND the ci-fail-closed tests can still patch it. (3) fast-lane-build.js dual-engine: meta.description is a pure literal, and the worktree_path early-return removed so the test-writer dispatch fires under the E2 stub. build.py re-synced deployed copies (hook-parity restored). Verified green: workflows 92, build_guards 30, dual-engine 28, ci-fail-closed 4, ci-changed 14, imports 23, hook 23; ruff clean. SKIP=check-product-truth-{validate,generate}: pre-existing unrelated index drift. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…shim (ci-fail-closed)
test_m2_ci_mode_returns_0_when_all_done_acs_are_eligible failed only in CI: the captured stdout ('no linked test found for BO-M2-PASS-001') showed the REAL verify_done_eligible ran despite the test mocking it — the prior call-site '_verify = verify_done_eligible if not None else _load_verify_done_eligible()' could reach the real function via the _load fallback, bypassing the patched module global. Fix: when the top-level done_proof import fails, bind verify_done_eligible to a lazy SHIM function (still a real, patchable module attribute) and call verify_done_eligible(...) directly at both CI call sites — so unittest.mock.patch('check_done_proof.verify_done_eligible') always takes effect and no code path bypasses it. Deployed copy re-synced via build.py (parity). Locally: ci-fail-closed 4 + imports 23 green. SKIP=check-product-truth-{validate,generate}: pre-existing unrelated index drift.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…oof gate + self-sufficient ci-fail-closed fixture
The required CI done-proof gate (dogfooding on its own PR) correctly flagged 14 PR-changed done ACs with no covers test — but they are documentation / prompt-convention ACs carrying test_required:false, which cannot have a '# covers:' test. Real design gap: the gate didn't honor test_required:false. Fix: check_all_done_acs + check_changed_done_acs now skip (treat eligible) any done AC whose YAML has test_required exactly False (absent/true still enforced); 5 new behavioral tests (test_done_proof_test_required_exemption.py). Behaviorally spot-checked: BO-2500a-4 (test_required:false docs AC) -> no violation; BO-2500a-3 (code AC) -> still enforced. Also fixed test_m2_ci_mode_returns_0 (CI-only failure: the mock didn't apply under full-suite module caching, real verify ran) by making its fixture self-sufficient with a real passing '# covers:' test — real-fixture-over-mock, the repo's own principle. 69 tests green; deployed copy re-synced (parity). SKIP=check-product-truth-{validate,generate}: pre-existing unrelated index drift.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Empty commit to re-trigger CI on ac-authoring/fast-lane — Actions did not start a run for 7a2cfb7 (known transient). No file changes. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This was referenced Jul 23, 2026
Merged
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Captures — as acceptance criteria only (no ticket files) — a lean "fast lane" build path for scoped, interactive feature work, plus the mechanical proof-of-done that makes it safe. Complements, does not replace, the heavy multi-agent pipeline (which stays for large/unattended/high-defect-cost drives).
Motivation: a 3-agent research pass found the heavy pipeline is ~1/3 inherent cost and ~2/3 accidental overhead (cold-context re-loading, deterministic steps run as agents, over-serialization, 3-command/3-worktree split), and that rework commits outnumber feature commits because LLM sign-off gates phantom-pass green while work is absent.
AC tree (all readiness: draft — this PR is the review)
BO-2400 — fast lane (L0) → BO-2400a lean two-agent loop · BO-2400b path selection · BO-2400c caching efficiency · BO-2400d telemetry. Depends on BO-2500 (hard edge).
BO-2500 — mechanical proof-of-done (L0) → BO-2500a test-backed requirements · BO-2500b unskippable CI gate · BO-2500c real-artifact fixtures · BO-2500d retire opinion-only review.
~60 ACs total: 2 L0, 8 L1, 39 L2/L3 testable behaviors, 13 documentation ACs. Enriched with assigned_agent (python-coder for deterministic gates/scripts, llm-expert for workflow/CI/template surface, doc agents for docs), it_requirements, contracts, and verified doc_links.
Notes / follow-ups
🤖 Generated with Claude Code