test-cycle-orchestratelisted

Drives one or more turns of the iterative test cycle defined in spec/project/test-cycle-foundation/ — case determination → execution → result analysis → code adaptation → re-execute — dispatching each phase to its capability (test-case-extractor and the per-tier test-generators for cases, quality-gate for execution, test-result-analyzer for analysis, test-code-adapter for the fix) and looping until an explicit exit condition holds. Enforces the cycle's integrity rules: a regression case before fixing a defect, a flake quarantined not retried-until-green, and the no-cheating invariant (never weaken/skip a test to force a pass). Invoke when the user asks to run the test cycle for a feature, drive a feature to green, or iterate determine-execute-analyse-fix. Don't use to run the gate once (use quality-gate), to scaffold tests without the loop (use a tier generator), or to classify results without acting (use test-result-analyzer). Supports resume per `spec/claude/resumable-work/`.
nolte/claude-shared · ★ 0 · Testing & QA · score 74

Install: claude install-skill nolte/claude-shared

# Test Cycle Orchestrate: $ARGUMENTS Drive the iterative test cycle for `$ARGUMENTS` (a feature, module, or failing case): determine cases → execute → analyse results → adapt code → re-execute, looping until an explicit exit condition holds. This skill **orchestrates**; each phase is performed by its capability, dispatched in turn, and the results flow back here so you decide whether to loop again. Governed by `spec/project/test-cycle-foundation/` (the loop, the inter-phase contracts, the exit conditions, and the no-cheating invariant) and the four phase specs it references. ## Why this is a skill, not an agent - **Orchestrator that chains other capabilities:** the cycle dispatches `test-case-extractor` / per-tier generators, `quality-gate`, `test-result-analyzer`, and `test-code-adapter` in a loop; the skill-orchestrates pattern (per `skill-vs-agent`) defaults the orchestrator to skill form. - **Mid-flow gating lives in the conversation:** the no-cheating decision, the "is this exit-ready or loop again" decision, and the two decisions that route a case back to a reviewable case change instead of a production edit — the test was wrong, and a relied-on permissive arrangement is what makes the production attribution doubtful — are per-turn user-visible gates an agent's fire-and-forget shape would lose. - **Output flows back into the main conversation:** each phase's result (the classification, the applied fix, the re-run verdict) surfaces so the operator can steer. - Counte