The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests content from external repositories and research databases (GitHub, arXiv, Zenodo) to verify facts stored in its memory files. This creates a surface where external data could potentially contain malicious instructions intended to influence the agent's behavior during the re-verification or archival process. The skill mitigates this by requiring human confirmation and comparing results against a secondary fact-checker.
[COMMAND_EXECUTION]: The skill employs bash for administrative tasks, including directory traversal, log generation using printf, and data integrity operations like creating compressed archives with tar. These operations are scoped to the agent's memory and metadata directories.
[EXTERNAL_DOWNLOADS]: To maintain memory hygiene, the skill fetches status updates and metadata from well-known services including GitHub (via the gh CLI) and scientific repositories (via WebFetch).

memory-hygiene