The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes standard developer tools such as git, gh (GitHub CLI), and AI provider CLIs (codex, claude) to analyze changes. These operations are managed via sanitized environments that exclude untrusted Git configuration and use absolute path overrides for binary resolution.
[DATA_EXPOSURE]: Test fixtures in tests/fixtures/typescript-sensitive-literals.ts include example hardcoded credentials. These are explicitly prefixed with 'FAKE' and serve as test cases to verify the reviewer's detection capabilities; they do not represent real secret leakage or exfiltration.
[DYNAMIC_EXECUTION]: The test suite employs standard Python testing patterns, using importlib and runpy to load and execute the local autoreview script for verification purposes. This does not involve the runtime execution of untrusted external scripts or code.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data from Git diffs and images. It mitigates injection risks through robust boundary controls, including partitioned processing, reviewer isolation, and restricted sandboxes to ensure that instructions embedded in analyzed data cannot influence the agent's operation.
[EXTERNAL_DOWNLOADS]: The skill interacts with official AI provider APIs (OpenAI and Anthropic) and well-known services like GitHub. It optionally requires the Pillow library for image processing, which is installed through standard, trusted package registries.

autoreview