The Agent Skills Directory

[SAFE]: The skill provides instructional material for an agent to evaluate its own performance across five axes (accuracy, completeness, clarity, actionability, conciseness). There are no malicious instructions, prompt injections, or exfiltration patterns.
[SAFE]: The provided Python script scripts/evaluate.py performs static analysis using regex patterns to score text. It does not execute the analyzed content, does not use dangerous functions like eval() or exec(), and does not make network requests.
[SAFE]: The hook integration documented in references/hook-integration.md suggests using standard agent harness features to provide user reminders via echo. It does not introduce persistence or privilege escalation risks.

agent-self-evaluation