The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-specified files and directories for quality review. If the code under review contains adversarial content (e.g., hidden instructions in comments or string literals), it could potentially influence the subagents' analysis or the automated 'fix' suggestions provided in Step 4.
Ingestion points: $ARGUMENTS provided by the user and the contents of files within the designated review scope (e.g., source code, CLAUDE.md configuration).
Boundary markers: The skill wraps arguments in <review_scope> tags and uses separate instruction files for subagents, though these do not fully mitigate adversarial input within analyzed data.
Capability inventory: The skill can write reports to the local file system, modify source code through automated fixes, and post comments to external platforms using gh or glab CLI tools.
Sanitization: Path inputs are validated by splitting on whitespace and checking existence, but the content of files processed by the review agents is not sanitized for prompt injection patterns.
[COMMAND_EXECUTION]: The skill executes a local shell script (detect-review-scope.sh) and utilizes the GitHub (gh) and GitLab (glab) CLI tools. While these operations are scoped to the skill's primary function of managing code reviews and pull request feedback, they represent powerful capabilities for interacting with both the local environment and remote repositories.

review