Tasks: Workflow Parity Fixes 988/989/991

Mission: workflow-parity-988-989-991-01KRKTT5 Branch: fix/workflow-parity-988-989-991 Plan: plan.md

Subtask Index

IDDescriptionWPParallel
T001Add ClaimablePreview dataclass + preview_claimable_wp() helperWP01[P]
T002Wire next --json payload to populate wp_id + selection_reasonWP01
T003Regression test: next --json exposes claimable wp_id (#988)WP01
T004Add LIGHTWEIGHT_REVIEW_MISSING_BASELINE + LEGACY_MISSION_DEAD_CODE_SKIP diagnostic codesWP02[P]
T005Update scan_dead_code() to fail-hard on modern missions with null baselineWP02
T006Regression test: lightweight review fails missing baseline on modern mission (#989)WP02
T007Extract run_review_artifact_consistency_preflight() helperWP03[P]
T008Invoke preflight in merge --dry-run path (human + JSON output)WP03
T009Regression test: merge --dry-run emits REJECTED_REVIEW_ARTIFACT_CONFLICT (#991)WP03

Work Packages

WP01 — next --json claimability parity (#988)

Goal: Make spec-kitty next --json expose the same WP that agent action implement would claim.

Priority: P1 — operator/agent loop trust.

Independent test: Given a mission with at least one claimable WP, spec-kitty next --json includes a non-null wp_id; given a mission with no claimable WPs, it includes a non-null selection_reason.

Included subtasks:

  • □ T001 Add ClaimablePreview dataclass + preview_claimable_wp() helper in src/specify_cli/next/
  • □ T002 Wire next --json payload (Decision.to_dict() + _print_decision) to invoke the helper and emit wp_id and selection_reason
  • □ T003 Add regression test tests/next/test_next_claimable_payload.py asserting the new payload shape

Implementation sketch: 1. Introduce ClaimablePreview and side-effect-free preview_claimable_wp(feature_dir, *, mission_meta) in src/specify_cli/next/discovery.py (or extend existing module). 2. Inside the next decision builder, when mission_state == "implement" and preview_step == "implement", call the helper and merge wp_id + selection_reason into the decision dict. 3. Add a regression test exercising two scenarios (claimable, not claimable).

Parallel opportunities: T001 is independent of T002/T003. T003 depends on T001+T002.

Dependencies: none.

Prompt: tasks/WP01-next-json-claimability.md

Estimated prompt size: ~280 lines.

WP02 — Lightweight review dead-code parity (#989)

Goal: Modern missions with baseline_merge_commit: null must fail lightweight review with a structured diagnostic rather than silently passing.

Priority: P1 — release-evidence integrity.

Independent test: Given a modern mission (with mission_id in meta.json) and baseline_merge_commit: null, spec-kitty review --mode lightweight exits non-zero with LIGHTWEIGHT_REVIEW_MISSING_BASELINE. Legacy missions (no mission_id) still pass with LEGACY_MISSION_DEAD_CODE_SKIP tag.

Included subtasks:

  • □ T004 Add LIGHTWEIGHT_REVIEW_MISSING_BASELINE and LEGACY_MISSION_DEAD_CODE_SKIP to src/specify_cli/cli/commands/review/_diagnostics.py
  • □ T005 Update scan_dead_code() in src/specify_cli/cli/commands/review/_dead_code.py to branch on mission_id presence + baseline_merge_commit
  • □ T006 Add regression test tests/specify_cli/cli/commands/review/test_dead_code_baseline.py

Implementation sketch: 1. Add the two diagnostic codes. 2. Modify scan_dead_code() to:

3. Wire the structured failure through the review verdict so the CLI exit code reflects the failure. 4. Add regression test covering all three branches.

  • If baseline_merge_commit populated → run existing scan (no change).
  • Elif mission_id populated → return structured failure with LIGHTWEIGHT_REVIEW_MISSING_BASELINE.
  • Else → preserve historical skip but tag with LEGACY_MISSION_DEAD_CODE_SKIP.

Parallel opportunities: T004 is independent of T005/T006. T005 depends on T004; T006 depends on T005.

Dependencies: none.

Prompt: tasks/WP02-lightweight-review-baseline.md

Estimated prompt size: ~320 lines.

WP03 — Merge dry-run review-artifact gate parity (#991)

Goal: spec-kitty merge --dry-run must invoke the same review-artifact consistency gate as real merge, surfacing REJECTED_REVIEW_ARTIFACT_CONFLICT in both human and JSON output.

Priority: P1 — release readiness signal trust.

Independent test: Given a mission with WP01 lane approved and a rejected latest review-cycle artifact, spec-kitty merge --mission <slug> --dry-run --json exits non-zero with REJECTED_REVIEW_ARTIFACT_CONFLICT in the JSON payload, and human output names the conflict.

Included subtasks:

  • □ T007 Extract run_review_artifact_consistency_preflight() helper used by both real merge and dry-run
  • □ T008 Invoke the preflight in the dry-run path (human + JSON output)
  • □ T009 Add regression test tests/specify_cli/cli/commands/test_merge_dry_run_review_artifact.py

Implementation sketch: 1. Extract the existing real-merge invocation of find_rejected_review_artifact_conflicts() into a single helper that returns a structured result. 2. Call the helper from the merge --dry-run path before computing the preview; on detection, emit the diagnostic to the chosen output (human or JSON) and exit non-zero. 3. Confirm tests/post_merge/test_review_artifact_consistency.py and tests/merge/test_merge_post_merge_invariant.py stay green. 4. Add regression test for the dry-run path.

Parallel opportunities: T007 is independent of T008/T009. T008 depends on T007; T009 depends on T008.

Dependencies: none.

Prompt: tasks/WP03-merge-dry-run-review-artifact.md

Estimated prompt size: ~320 lines.

MVP Scope

Each of WP01, WP02, WP03 is independently shippable. Recommended order: WP03 → WP01 → WP02 (per start-here.md execution order — WP03 is most concrete, WP01 needs payload understanding, WP02 has the most policy nuance).

Parallelization

All three WPs touch disjoint code paths:

  • WP01 owns src/specify_cli/next/ + src/specify_cli/cli/commands/next_cmd.py payload edits.
  • WP02 owns src/specify_cli/cli/commands/review/.
  • WP03 owns src/specify_cli/cli/commands/merge.py + src/specify_cli/post_merge/review_artifact_consistency.py.

They can run in parallel across three lanes if needed; in this mission they are executed sequentially by a single agent.