The Agent Skills Directory

[COMMAND_EXECUTION]: The backup procedure in Step 2 of SKILL.md executes a Bash command using variable interpolation: cp "${FILE_PATH}" "${backup_path}". If the FILE_PATH parameter provided to the skill contains shell metacharacters (such as semicolons or backticks), it could allow for unauthorized command execution on the host system.\n- [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection attacks because it ingests untrusted document content into an agent's reasoning loop.\n
Ingestion points: The contents of the document specified by FILE_PATH are read and directly inserted into the persona analysis agent's prompt in Step 3.\n
Boundary markers: There are no explicit delimiters or instructions to the agent to treat the document content as untrusted or to ignore embedded instructions within that content.\n
Capability inventory: The skill possesses powerful capabilities including Bash, Edit, and Write, allowing it to modify the file system and execute shell commands.\n
Sanitization: The skill does not perform any sanitization or filtering of the document content before it is processed by the AI persona agents.\n- [PROMPT_INJECTION]: The AUTO_IMPROVE workflow in Step 5 automatically applies file edits using the Edit tool based on recommendations generated from the analysis of the document. A malicious document could contain instructions that trick the agents into recommending destructive or unauthorized changes, which the skill would then apply without human review.

copy-reader