The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes standard version control commands to identify and extract code changes for analysis.
SKILL.md: Executes git rev-parse and git log to determine the start and end commit identifiers (SHAs) for the review range.
code-reviewer.md: Instructs the subagent to execute git diff --stat and git diff using the provided SHAs to inspect the modified code.
[PROMPT_INJECTION]: The skill has a surface for indirect prompt injection because the subagent processes untrusted external data (code changes and user-provided descriptions).
Ingestion points: External data enters the context via the git diff command output and the {WHAT_WAS_IMPLEMENTED}, {PLAN_OR_REQUIREMENTS}, and {DESCRIPTION} placeholders in code-reviewer.md.
Boundary markers: The template uses markdown headers (e.g., ## Requirements/Plan) to separate sections, but lacks explicit instructions to ignore embedded commands within the code being reviewed.
Capability inventory: The subagent has the capability to execute git commands as specified in the template.
Sanitization: No explicit sanitization or validation of the commit range or placeholder content is implemented.
Assessment: This is a common and necessary pattern for automated code review tools; the risk is considered low within the intended scope of development workflows.

requesting-code-review