The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides utility scripts in references/memory-hygiene.md and references/ab-test-harness.md using bash and python. These commands are used for legitimate purposes: cleaning up internal agent memory files (episodic logs, fact stores, and indices) and analyzing execution transcripts to measure tool usage.
[DATA_EXFILTRATION]: The skill involves accessing internal configuration and memory files under the .claude/ directory. This access is necessary for the skill's stated purpose of "memory hygiene" and optimizing agent context. No evidence of data being sent to external or unauthorized destinations was found.
[SAFE]: The skill contains well-documented instructions for improving agent accuracy and reducing token costs. It explicitly warns against anti-patterns like "hoarding hot context" or "burying must-apply rules in cold storage," and provides a robust A/B testing framework for validating changes.

tune-skills-and-agents