The Agent Skills Directory

[SAFE]: The skill implements significant safety controls for processing local files.
The snapshot logic in scripts/snapshot.py is explicitly designed to skip symlinks, preventing potential path traversal or sensitive data disclosure if a target skill contains links to files like ~/.ssh/id_rsa.
The workflow instructions in SKILL.md require the agent to stop and request confirmation if a file path contains shell metacharacters, effectively mitigating command injection risks.
The verification process uses deterministic logic in scripts/verify_rewrite.py to ensure that no instructions, rules, or literals are lost or modified during optimization. This script also enforces that no new text is added to the optimized skill, which serves as a strong defense against indirect prompt injection from the source material being processed.
All logic is contained in local scripts using standard Python libraries, with no external network calls or untrusted code execution detected.

skill-optimizer