Tasks: CI Scoping Gate Reliability
Mission: ci-scoping-gate-reliability-01KZP80D | Branch: fix/ci-scoping-gate-reliability Spec: spec.md | Plan: plan.md Authoritative squad remediation: tracer-squad-findings.md — B1/M1-M5/N1-N2 are binding.
Two independent work packages, one per blocking CI gate. They touch disjoint files (ci-quality.yml + arch guards vs docs-freshness.yml + docs scripts/tests), so they run as two parallel lanes.
Subtask Index
| ID | Description | WP | Parallel |
|---|---|---|---|
| T001 | Discrete corpus globs in ci-quality.yml pull_request.paths and push.paths (B1, no braces) | WP01 | |
| T002 | corpus change-group: changes.outputs.corpus row + corpus: dorny filter (narrow globs) | WP01 | |
| T003 | fast-tests-corpus blocking job (modeled on fast-tests-docs, -m corpus, not whole dirs) | WP01 | |
| T004 | Wire the gate: JOB_GROUPS["fast-tests-corpus"]=["corpus"] + quality-gate.needs edge | WP01 | |
| T005 | Register corpus marker in pytest.ini + apply module-level pytestmark to corpus readers | WP01 | |
| T006 | Co-evolve test_ci_quality_path_filters.py + test_ci_collection_completeness.py | WP01 | |
| T007 | Corpus-trigger completeness invariant (M4): every corpus-read path matched by the globs | WP01 | |
| T008 | Diff-scope mode in relative_link_fixer.py (fail-closed base-ref, M5) | WP02 | [P] |
| T009 | Diff-scope mode in related_validator.py --strict (same fail-closed base-ref, M5) | WP02 | [P] |
| T010 | docs-freshness.yml: PR runs diff-scoped from base.sha; retain unfiltered push:main backstop | WP02 | |
| T011 | Co-evolve test_docs_freshness_invariant.py (keep backstop + PR-allowlist shape) | WP02 | |
| T012 | Co-evolve test_rulers_blocking.py (pass base/changed-set; seeded-violation RED stays green) | WP02 |
The [P] marker indicates parallelism, not status. Record completion with spec-kitty agent tasks mark-status <Txxx> --status done.
WP01 — Corpus data triggers and gates the blocking corpus suite (#3008)
Prompt: tasks/WP01-corpus-gate-trigger-and-blocking-job.md Priority: P1 (US1 + US2) | Dependencies: none | Requirements: FR-001, FR-002, FR-003, NFR-001, NFR-002, NFR-003, NFR-004, C-001, C-003 | Estimated prompt size: ~420 lines
Goal: A data-only PR (packs/, narrow kitty-specs planning files, .kittify corpus config) triggers the quality workflow (closes Gate 0) and runs the corpus-reading suites as a blocking** job feeding quality-gate (closes Gate 1) — without firing on status.events.jsonl/notes/trace churn and without double-running whole already-covered directories.
Independent test: Path-filter tests prove a packs/built-in/**-only diff selects the corpus group and the fast-tests-corpus job; a kitty-specs/<m>/status.events.jsonl-only diff does NOT.
Included subtasks: T001, T002, T003, T004, T005, T006, T007
Implementation sketch: 1. Extend the two trigger allowlists with discrete corpus globs (B1). 2. Register the corpus change-group exactly like the docs precedent (M2) — five atomic edits. 3. Add the fast-tests-corpus blocking job (-m corpus, M1) and wire it into quality-gate. 4. Mark the corpus-reading modules and co-evolve the arch invariants + a new completeness guard.
Risks: brace-glob inertness (B1); double-run of whole dirs (M1 — bound with the marker); touching ci_topology_census.json/src_backed_groups/the unmatched loop reds test_ci_topology_worklist (M2); a missing quality-gate.needs edge reds test_suite_jobs_gate_blocking.py (N2).
WP02 — Docs dead-link gate is scoped to the PR's own diff (#3147)
Prompt: tasks/WP02-docs-deadlink-diff-scope.md Priority: P2 (US3) | Dependencies: none | Requirements: FR-004, FR-005, C-002 | Estimated prompt size: ~320 lines
Goal: Scope the two blocking whole-tree dead-link checks (relative_link_fixer.py --check, related_validator.py --strict) to the PR's changed files, so unrelated pre-existing broken links no longer fail a docs PR — while the unfiltered push:main whole-tree run is retained as the non-blocking rot backstop (FR-005). The check still BITES on a link the PR itself breaks.
Independent test: A docs PR whose own diff has no broken links passes even when an untouched file elsewhere has one; introduce a broken link in a changed file → it FAILS; the push:main scan still reports the untouched link.
Included subtasks: T008, T009, T010, T011, T012
Implementation sketch: 1. Add a fail-closed diff-scope mode to both scripts (base-ref → changed docs files; empty set errors). 2. Update docs-freshness.yml to run the two gates diff-scoped on pull_request (from base.sha, fetch-depth: 0) while leaving the unfiltered push:main whole-tree run intact. 3. Co-evolve the two docs invariants.
Risks: false-green if the base-ref/changed-set is empty and the check passes trivially — MUST fail-closed (M5); docs-freshness is NOT a required check — do not wire it into quality-gate.needs (N1); the push:main backstop must survive (M3/C-002).
MVP scope
WP01 (#3008) is the MVP: it closes a genuine correctness hole in the release gate (corpus regressions shipping invisibly). WP02 (#3147) removes per-PR friction and is lower-stakes (rot is still caught by the retained whole-tree scan).