The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted Python source code and diff data provided by the user. It has the capability to invoke shell commands, including the pytest test runner, which could theoretically execute malicious logic if it were embedded within the code or tests being reviewed. However, the risk is mitigated by the skill's primary design, which focuses on providing an evidence-based review framework.
Ingestion points: Python source files and git diff output identified in SKILL.md Step 1.
Boundary markers: The skill implements a mandatory 'Review Verification Protocol' (references/review-verification-protocol.md) and five 'Hard Gates' (SKILL.md) that require the agent to ground all findings in objective evidence, linter output, and exact source code quotes.
Capability inventory: Execution of git, grep, ruff, mypy, and pytest shell commands as specified in SKILL.md.
Sanitization: Findings must be verified against official documentation and tool outputs, with a strict 'Anti-confabulation' gate that forces the agent to echo the exact artifact being judged.
[COMMAND_EXECUTION]: The skill executes several standard development tools via the shell (git diff, ruff check, mypy, pytest) to perform its review functions. These commands are used for their intended purposes within a development environment and do not show signs of malicious intent or unauthorized access.

do-review-python