The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The build_docx.py script implements a preprocessor that resolves  sentinels in markdown drafts to inline their contents.
Ingestion points: The script reads and processes all .md files in the drafts/ directory.
Boundary markers: The inclusion mechanism lacks boundary markers to prevent the agent from interpreting or executing instructions contained within the included files.
Capability inventory: The script has the ability to read local files via path.read_text() and executes system commands through subprocess.run (e.g., pandoc, pdftotext).
Sanitization: While the preprocessor requires absolute paths and expands ~, it does not restrict access to specific directories, allowing it to potentially read sensitive files such as SSH keys or environment variables if a malicious include directive is provided in the draft.
[COMMAND_EXECUTION]: The skill uses subprocess.run to invoke pandoc for document building and pdftotext for widow detection. These represent significant system capabilities that rely on the integrity of the project path and file inputs.
[DYNAMIC_EXECUTION]: The build_docx.py script dynamically modifies the Python sys.path to import modules from relative paths outside of the skill's specific directory (e.g., a sibling bluebook-audit directory). This creates a runtime dependency on a specific external folder structure on the host system.

law-review-docx