Fix The Suite

Fix The Suite

Replaces the pass-1 / pass-2 / pass-3 cycle (where each pass deletes 16 tests then bails out at the coverage gate) with one workflow that runs all the right skills in the right order.

Auto-invocation triggers

• Test count is >2× the proportionality target for the app type
• Suite runtime >5 minutes
• User reports "tests are slow", "too many tests", "test cleanup didn't work"
• After any cleanup pass that reported "needs further work / out of scope"

Workflow

Phase 1 — Diagnose (always)

Invoke test-health for the full picture: count vs proportionality, coverage vs threshold, runtime, slowest tests, dead weight, mocked-SUT count, flake estimate.

If verdict is HEALTHY → exit; nothing to fix.

Phase 2 — Gate compatibility check (always)

Invoke config-drift-detect scoped to test config (jest/vitest thresholds). If a CRITICAL gate conflict surfaces (e.g., 99% functions on a 3k-LOC project), resolve it before pruning:

• Surface the conflict per the test-cleanup Step 1.5 protocol
• Get explicit user choice: lower gate / exclude files / commit to integration tests
• Do not start Phase 3 until resolution is recorded

Phase 3 — Prune + replace (always)

Invoke test-cleanup with the resolution from Phase 2. The skill already chains internally: skipped tests → whole files → waste patterns → batched coverage check → integration test replacement (it.each) → consolidation.

Phase 4 — Validate survivors (always)

Invoke mutation-test per-file or scoped to changed files. If mutation score <60% on critical modules: loop back to Phase 3 to write real-behavior tests for the survivor mutants.

Phase 5 — Capture (mandatory if Phase 2 changed gates)

Invoke adr-write to record:

• The gate change (if any) and why
• The cleanup outcome
• The proportionality reasoning Without an ADR, the next person will revert the gate change.

Phase 6 — Sync (always)

Invoke docs-sync to ensure any modified standards/skills propagate.

Reconciliation

Final report:

FIX THE SUITE — <repo>
  Diagnose:   tests=N coverage=X% runtime=Ts → BLOATED
  Gates:      <conflicts found, resolution>
  Cleanup:    N → M tests (-X%), runtime -Y%, coverage maintained at X%
  Mutation:   score <Z>% on changed files
  ADR:        <path> (if gate changed)

Outputs / Evidence

• test-health report (before)
• config-drift verdict + resolution
• test-cleanup report (deletions, integration tests written)
• mutation score
• ADR path if applicable
• Final coverage proof from CI

Failure / Stop Conditions

• Phase 2 conflict cannot be resolved (user declines all options) → stop with clear "this suite cannot be efficiently cleaned without one of: A/B/C"
• Phase 4 mutation score collapses (<40%) after cleanup → revert the cleanup, the deletions removed real protection
• Never skip Phase 2 — it's the failure mode this whole workflow exists to prevent