The Agent Skills Directory

[COMMAND_EXECUTION] (SAFE): The extract-skill.sh script scaffolds new skill directories and files with strict regex validation on the skill name input, preventing path traversal.
[DATA_EXFILTRATION] (SAFE): The error-detector.sh script reads tool output for local error detection without external network operations or data transmission.
[PROMPT_INJECTION] (SAFE): The skill uses hooks to inject hardcoded reminder strings into the session context, which does not involve processing or echoing untrusted data.

self-improvement