The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input in the form of symptoms, error messages, and log files provided via the $ARGUMENTS variable or the Problem Intake phase. This creates a surface for indirect prompt injection where malicious instructions could be embedded in the data being debugged.
Ingestion points: Data enters through the $ARGUMENTS parameter and text intake templates for error messages and logs in SKILL.md.
Boundary markers: The skill uses markdown code blocks and text delimiters to separate input data from instructions, though these do not prevent adversarial injection.
Capability inventory: The skill uses shell commands via uv for testing and linting, and implies file modification to implement fixes.
Sanitization: No explicit sanitization or filtering of input logs or error messages is specified.
[COMMAND_EXECUTION]: The skill recommends executing shell commands to verify bug fixes using the uv package manager. These are standard development tasks for testing and linting code.
Evidence: The 'Checks After Fix' section in SKILL.md lists uv run prek run, uv run ruff check, and uv run pytest -v as verification steps.

debug