The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes dynamic context injection (the !command syntax) within SKILL.md to execute several shell commands at load time, specifically: git status, git branch --show-current, git merge-base, and various git diff variants. These commands are used legitimately to populate the agent's context with the current state of the repository for review purposes.
[PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection because it ingests untrusted data from the user's codebase via code diffs.
Ingestion points: Content is pulled into SKILL.md via !git diff and !git status commands.
Boundary markers: The ingested content is wrapped in XML-style tags (e.g., <branch_diff>, <staged_diff>). However, there are no explicit instructions telling the agent to ignore potential instructions embedded within those diffs.
Capability inventory: The skill uses the Task tool to launch parallel reviewers. If a reviewer is successfully injected, it could influence the final report or subsequent task execution.
Sanitization: No sanitization or filtering of the diff content is performed before interpolation into the prompt.

check