The Agent Skills Directory

[SAFE]: The skill implements a robust workflow for validating Python scripts within a controlled workspace. It defines clear boundaries, such as prohibiting the modification of raw data files and forbidding the fabrication of results.
[COMMAND_EXECUTION]: The skill instructions involve checking if Python scripts are runnable and identifies imports and dependencies. It does not perform arbitrary shell command execution; instead, it provides recommendations for fixing path errors or missing dependencies using standard scientific Python practices.
[INDIRECT_PROMPT_INJECTION]: The skill processes external content including generated Python code and method plans. While it lacks formal sanitization for these inputs, the risk is mitigated by its analytical focus and the requirement to surface semantic logic (e.g., mathematical constraints) for human review rather than applying automatic corrections.
Ingestion points: Reads files from workspace/code/python/ and workspace/problem/method-selector/.
Boundary markers: None explicitly defined in the file ingestion process.
Capability inventory: Limited to reading workspace files, writing review reports, and performing surgical code fixes (e.g., path corrections or missing imports).
Sanitization: Relies on reporting findings to a mandatory review file for human oversight.

python-code-reviewer