The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/audit.py script uses subprocess.run to execute an external Python tool doc_nav_lint.py. The path to this tool is dynamically resolved to a sibling directory (doc-init), creating a dependency on code outside the skill's own package.
[REMOTE_CODE_EXECUTION]: The documentation maintenance workflow in references/maintenance-workflow.md instructs the agent to execute multiple Python scripts from an external <DOC_INIT_DIR> location, including discover_global_instruction_files.py and insert_doc_governance.py, which are used to modify global AI instructions.
[DATA_EXFILTRATION]: The scripts/audit.py script includes a --save-metrics argument that allows writing JSON audit data to a user-specified file path. This capability could be exploited to overwrite sensitive files (e.g., shell profiles or configuration files) if the agent is directed to a malicious destination path.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to ingest, audit, and rewrite documentation, which establishes an attack surface for indirect prompt injection.
Ingestion points: scripts/audit.py and the compression logic in references/maintenance-workflow.md read all markdown files within the docs/ and specs/ directories.
Boundary markers: The skill utilizes HTML comment markers () for content preservation but does not implement security boundaries to prevent the AI from following instructions embedded in the processed documentation.
Capability inventory: The agent is empowered to modify files using specialized tools and shell commands, while the audit script itself has file-writing permissions.
Sanitization: The skill does not perform sanitization or filtering of documentation content before it is presented to and processed by the agent.

doc-compact