The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input from source code changes, commit messages, and external documentation.
Ingestion points: Git diff outputs, commit logs, and specification files located in directories like docs/, specs/, or .scratch/.
Boundary markers: The skill employs a security-positive pattern by writing review materials to a temporary file rather than injecting raw diff data directly into the agent's prompt context.
Capability inventory: The skill uses standard git read operations and local file system access.
Sanitization: The instructions include specific directives for sub-agents to ignore extraneous context and focus on specific review criteria.
[COMMAND_EXECUTION]: The skill utilizes standard version control commands (git diff, git log, git rev-parse, git show) to analyze the repository state. These commands are consistent with the skill's purpose and are used for data collection rather than system modification.
[SAFE]: The skill includes explicit instructions to identify security vulnerabilities such as hardcoded credentials, authorization flaws, and missing input sanitization in the code being reviewed, which enhances the overall security posture of the development workflow.

code-review