crew-review

You can audit autonomous agent configurations by running an adversarial audit that stress-tests agent charters, skills, and memory against recent task outcomes to identify stale data and operational failures.

Adversarial self-audit is a maintenance process where a separate agent stress-tests your crew's charters, skills, and memory corpus against recent outcomes to identify contradictory principles and degraded system performance.

To fix contradictory principles and stale memory, you apply automated maintenance that automatically resolves stale memory data while staging gated configuration changes for human approval.

Yes, performing an agent configuration audit requires access to the .claude directory and the memory corpus to validate configuration integrity and propose gated self-modifications.

Yes, you can turn recurring manual procedures into reusable skills through capability distillation, which provides a structured path to crystallize repeated manual workflows into native skills.

The best way to prevent configuration drift is to perform regular adversarial audits comparing recent task outcomes against established principles and agent definitions to identify and resolve discrepancies.