The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted manuscript content which could potentially contain instructions intended to influence the agent's behavior during the audit process. \n- Ingestion points: Manuscript text files read and analyzed by the agent and the included check_figure_refs.py script. \n- Boundary markers: Absent; there are no specific instructions to treat manuscript content as data distinct from instructions. \n- Capability inventory: The skill performs read-only analysis and lacks capabilities for network communication, persistent file modification, or command execution beyond the provided script. \n- Sanitization: Absent; the manuscript content is processed as raw text without validation or filtering.

submission-audit