The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by interpolating untrusted input into a subagent prompt.
Ingestion points: Untrusted data enters the agent context via the [DESCRIPTION] and [PLAN_OR_REQUIREMENTS] placeholders in code-reviewer.md.
Boundary markers: Data is delimited by Markdown headers, but there is no specific "ignore embedded instructions" warning to prevent the subagent from following commands within the review material.
Capability inventory: The subagent can execute git diff, git show, git log, and git worktree as defined in code-reviewer.md.
Sanitization: No sanitization or escaping of the interpolated variables is present.
[COMMAND_EXECUTION]: The skill utilizes shell-based commands for its core functionality.
Evidence: Use of git rev-parse, git log, git diff, and git worktree across SKILL.md and code-reviewer.md to analyze repository state and code history.
[DYNAMIC_EXECUTION]: The skill generates subagent instructions at runtime by interpolating text into a predefined template.
Evidence: The use of [DESCRIPTION], [PLAN_OR_REQUIREMENTS], [BASE_SHA], and [HEAD_SHA] placeholders in code-reviewer.md creates a dynamic prompt for the general-purpose subagent based on the current task context.

requesting-code-review