adaclisted

# ADAC ## Purpose Use Acceptance-Driven Agentic Coding (ADAC) to make AI coding work verifiable before it becomes large. The core rule is: `Context + Spec + Evals + Gates` before broad implementation. ADAC is not ordinary TDD. Prefer behavior checks, regression checks, performance/pressure evidence, observability, and human decision gates over coverage targets. ## Default Workflow 1. **Classify risk**: assign risk class A, B, C, or D. - A: local isolated change. - B: shared or user-facing change. - C: critical path, concurrency, resource lifecycle, performance, release path, or broad integration. - D: security, privacy, irreversible data, payment, legal, or high-impact rollout. 2. **Build Context Pack**: list verified facts, source paths, docs, tests, incidents, metrics, logs, assumptions, and known unknowns. 3. **Define Spec Boundary**: write target behavior, non-goals, changed/stable interfaces, ownership boundaries, compatibility, and rollback/fallback. 4. **Create Acceptance Matrix**: convert real risks into checkable AC items with method, evidence, and owner/gate. 5. **Plan Agent Work**: decompose implementation into small steps; tie each step to acceptance items. 6. **Execute and Verify**: implement, test, profile, inspect logs, collect evidence, and update the matrix when new risk appears. 7. **Run Human Gates**: stop for human approval where judgment is required. 8. **Assetize**: preserve reusable context, checks, scripts, runbooks, specs, or skill u