The Agent Skills Directory

[COMMAND_EXECUTION]: The skill automates initialization and validation by executing local Python scripts such as init_skill.py and quick_validate.py. These scripts are expected to reside within the user's local environment (e.g., $CODEX_HOME).
[DATA_EXFILTRATION]: The instructions include robust privacy guidelines that forbid the use of absolute paths or usernames in code or documentation. It provides a dedicated script, check_path_privacy.py, to audit and replace identifying information with placeholders like $HOME or $CODEX_HOME before files are committed.
[PROMPT_INJECTION]: As a tool designed to generate new instructions for future agent sessions, this skill possesses an inherent indirect prompt injection surface (Category 8).
Ingestion points: User-provided goals, examples, and triggers are used to populate the content of new SKILL.md files.
Boundary markers: There are no explicit instructions to wrap or sanitize user-provided examples when they are interpolated into the generated skill body.
Capability inventory: The skill has the ability to write to the file system and execute local scripts for validation.
Sanitization: No content validation or escaping mechanisms are mentioned for the user-supplied input during the generation process.

ccf-skill-forger