The Agent Skills Directory

[SAFE]: The skill implements secure file system interactions by restricting state discovery to the current Git repository or working directory and explicitly forbidding the following of symlinks to prevent directory traversal attacks.
[SAFE]: Robust instructions are included to mitigate indirect prompt injection by directing the agent to treat local state files as data rather than instructions, effectively decoupling retrieved content from the agent's control flow.
[SAFE]: The skill explicitly prohibits the storage of secrets or transcripts and does not use any network tools (curl, wget) or remote code execution patterns. It relies on standard local file tools for its pedagogical state management.

learn