performance-goal

Automate latency optimization by running an evaluator-gated loop that repeatedly executes performance benchmarks and only proceeds when a declared pass/fail contract is satisfied. This turns optimization into a repeatable, evidence-driven process instead of a one-off review.

Track performance benchmark state by preserving durable artifacts under the `.omagy/goals/performance/<slug>/` directory. This ensures your optimization progress and evaluator checkpoints remain auditable across multiple workflow turns and sessions.

Set up a performance goal by defining a clear objective, an evaluator command, and an evaluator contract to serve as the optimization target and source of truth. This approach directly applies to memory, throughput, and speed targets.

Yes, you can use an evaluator loop to gate performance completion. It prevents final completion until a passing checkpoint exists and a completion audit confirms the benchmark objective is successfully achieved and recorded.

Yes, performance optimization supports optional host-goal parity via the `--host-goal-json` flag. This allows you to synchronize internal evaluator contracts with external benchmark targets during the optimization workflow.

Performance optimization workflows fail to complete without a passing checkpoint because the system enforces pass-only completion gating. It requires verification evidence from the evaluator command before marking the objective done.