What problem does it solve? Testing an AI agent that controls Ableton Live is hard because natural-language prompts produce nondeterministic behavior and assistant prose cannot be trusted as proof of success. This Skill provides a deterministic harness that runs reviewed scenario manifests against a real, runner-owned Live process and verifies exact project mutations through bridge reads and postcondition assertions. ## Core Features & Use Cases - Deterministic scenario execution: Run pnpm live:agent-smoke to execute reviewed scenario manifests with explicit tool allowlists, budgets, and cleanup assertions against a fresh Live Set. - Failure triage and layered reproduction: Classify failures across agent planning, approval policy, tool, bridge/protocol, Remote Script/LOM, assertion, timeout, and cleanup layers, then reproduce at the narrowest deterministic layer. - Coverage expansion guidance: Add new real-Live scenarios only after unit, protocol, bridge, tool, and workflow regression tests exist, with strict safety rules such as never using --approve-all or broadening allowlists to force a pass. - Use Case: After modifying the Python Remote Script, run the smoke suite so the harness reinstalls the script, launches a fresh Live process, and verifies scenarios like creating an 808 track or writing four-on-the-floor MIDI notes with exact note-content assertions. ## Quick Start Ask the agent to run the Ableton integration smoke suite with pnpm live:agent-smoke and triage the first failing scenario using the failure-triage reference.
