The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by reading external file content and passing it directly to a sub-agent for processing.
Ingestion points: Contents of files specified in $ARGUMENTS, the output of git diff commands, and the local .agents/review-perspectives.md configuration file.
Boundary markers: Absent. The instructions do not specify the use of delimiters or 'ignore' instructions when providing data to the sub-agent.
Capability inventory: The skill invokes a general-purpose sub-agent using the Agent tool, which possesses its own set of capabilities.
Sanitization: No sanitization, escaping, or validation of the ingested data is performed before interpolation into the sub-agent's prompt.
[COMMAND_EXECUTION]: The skill executes local git commands to retrieve data for the review process.
Evidence: Uses git merge-base and git diff to extract code changes when 'diff' is provided as an argument.

self-review