The Agent Skills Directory

[COMMAND_EXECUTION]: The testing framework (scripts/test_file_io.py) utilizes subprocess.run to invoke the worker script (file_io.py) for automated feature validation and error handling verification.
[INDIRECT_PROMPT_INJECTION]: The skill includes functionality to read files, which presents an inherent surface for external data to be loaded into the agent's context.
Ingestion points: File contents are read via the do_read function in scripts/file_io.py and returned in the JSON response.
Boundary markers: Output is encapsulated within a structured JSON object (ok, content, path, etc.) to differentiate it from instructions.
Capability inventory: The skill allows creating, writing, appending, and reading files; it can also create parent directories and perform atomic-style writes with rollbacks.
Sanitization: Content is handled verbatim with minor hygiene (newline normalization and BOM stripping). Crucially, directory traversal is prevented through os.path.commonpath validation against the working root, which is active by default.

file-io