The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses local shell commands like ls and git to manage and verify the presence of skills in project and global configuration directories. This is legitimate administrative behavior.
[DATA_EXFILTRATION]: Provides clear guidance to the agent to avoid capturing passwords or tokens, instead directing it to note where those secrets are stored. This mitigates the risk of credential leakage in shared repositories.
[PROMPT_INJECTION]: The skill processes session history to extract procedures, which creates a surface for indirect instructions. It utilizes sub-agent forks with structured prompts to create boundaries and specifies that content should be generalized and secrets excluded, effectively mitigating the risk of persisting malicious instructions.

self-learning