The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/diff_surgeon.py script executes git diff via subprocess.run to analyze staged or specified code changes. The implementation follows security best practices by using an argument list instead of a single shell string and avoids shell=True, which prevents shell command injection. This is standard and safe functionality for a code analysis tool interacting with version control.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted external data (such as code diffs and project plans) for heuristic analysis, which constitutes a theoretical vulnerability surface.
Ingestion points: Data is read from stdin or specific file paths in scripts/assumption_linter.py, scripts/goal_verifier.py, scripts/complexity_checker.py, and scripts/diff_surgeon.py.
Boundary markers: None; analyzed content is processed directly without delimiters to separate it from the linter's execution logic.
Capability inventory: The skill executes the git binary and reads local text files.
Sanitization: Input text is processed solely through regular expression matching for pattern detection; the skill does not dynamically evaluate, compile, or execute any part of the ingested content.

karpathy-coder