# Agent workflow controls lab

Run `python agent_cases.py` from the repository root. The lab executes deterministic Python state machines and tool simulators for termination, argument validation, approval, checkpoints, idempotency, retry/time budgets, cancellation, memory separation, delegation bounds, trajectory checks and partial completion.

No LLM, orchestration framework, external agent, network service or external write runs. Multi-agent and framework comparison outcomes are authored fixtures. The verifier’s passing state establishes only that the local control examples satisfy their declared invariants.

The verifier passed all 20 checks on 20 September 2026 and saved `agent-verification.json`. Human editorial review and live framework comparison remain pending.