skill-comply

You verify agent behavior by comparing expected behavioral sequences from markdown specs against actual JSONL tool-call traces. This compliance measurement uses LLM-based classification and deterministic temporal ordering checks to determine if agents follow rules.

Yes, you can test prompt independence by running scenario stress tests at multiple strictness levels. This process applies supportive to competing prompts to measure whether agent behavior and tool-call traces change under pressure.

Measuring trace compliance involves generating expected behavioral sequences from markdown specifications and comparing them to observed tool calls. It applies LLM-based classification to evaluate whether actual agent behavior aligns with defined rules.

Yes, compliance testing works directly with markdown-based agent instructions. It generates expected behavioral sequences from markdown specs and compares them against observed tool calls to ensure workflow verification and quality maintenance.

You generate a compliance report by capturing JSONL tool-call traces during scenario testing and evaluating them against markdown specs. The resulting report includes specs, prompts, timelines, and compliance scores for maintenance review.