The Agent Skills Directory

[COMMAND_EXECUTION]: The skill and its associated test harness scripts (autoreview, test-review-harness.py, and shell wrappers) execute system commands such as git, gh, and various AI model CLI tools. This is a primary requirement for a code review tool that must interact with repositories and external models.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted code diffs and PR content, which is a known attack surface for indirect prompt injection. However, the instructions in SKILL.md explicitly mandate isolation techniques, such as running engines in empty temporary workspaces, using 'safe modes' for external CLIs, and disabling project-local configuration loading to prevent the model from being influenced by malicious instructions in the code under review.
[DYNAMIC_EXECUTION]: The test suite (scripts/autoreview_test.py) uses Python's importlib and runpy to dynamically load and execute the main autoreview script for unit testing. While this pattern is often associated with obfuscation, here it is used for the legitimate purpose of testing a tool that is distributed as an extensionless executable script.
[CREDENTIALS_UNSAFE]: The file tests/fixtures/typescript-sensitive-literals.ts contains several hardcoded credential-like strings (e.g., sk-proj-FAKE..., ghp_FAKE...). These are clearly identified as mock data using 'FAKE' prefixes and are used specifically as fixtures to test the skill's integration with the TruffleHog secret scanner, rather than being active or leaked credentials.

autoreview