skill-comply

Measuring agent compliance involves automating the verification of whether coding agents follow defined skills by generating executable specs, executing scenarios, and classifying tool-call traces to compute a deterministic compliance score.

Automating compliance reporting for LLM agent behavior requires running agent sessions, collecting tool-call traces, and classifying them against expected steps to generate self-contained Markdown reports suitable for auditing.

Generating a compliance spec from skill content is automated by extracting defined rules and agent definitions to produce an executable YAML spec, which then drives scenario prompts for testing agent behavior.

YAML specs and trace data are required to compute a compliance score, as the system needs defined rules to classify tool calls and needs a compliant environment to execute representative scenarios.

Testing agent behavior with varying prompt strictness is supported by producing three scenario prompts with decreasing strictness, allowing you to observe how agents perform under different instruction constraints.

Compliance reports are delivered as self-contained Markdown files, providing a clear timeline and deterministic metrics suitable for auditing agent behavior and tool-call classification.