I use DProvenanceKit to catch AI agent failures that traditional output tests can miss, such as skipped verification steps or unexpected changes in execution paths. It’s designed for developers who want to catch workflow regressions before deployment.

1. Record: DProvenanceKit captures the agent’s instrumented execution steps.
2. Compare: It compares each new run against a known-good baseline.
3. Detect: It identifies missing steps, unexpected execution-path changes, and violations of configured rules.
4. Prevent: A GitHub Actions integration can automatically fail CI when a regression violates the configured policy, preventing the change from being merged or deployed when that check is required.
5. Try it: Run pip install dprovenancekit, followed by dpk demo. No API key or paid AI service is required.

The result: Instead of discovering that an AI agent silently skipped a critical step after deployment, developers can catch the regression automatically during testing.