The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses dynamic context injection (!command syntax) to run local git commands such as git status and git diff at load time. This is a legitimate use of the feature to gather project context for the AI agent.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it processes untrusted source code content through git diffs.
Ingestion points: Untrusted diff data from the local repository is loaded into the context using dynamic execution placeholders.
Boundary markers: The skill attempts to structure this untrusted data using XML-style tags like <staged_diff> and <branch_diff>.
Capability inventory: The skill has the capability to launch multiple sub-agents via the Task tool to process the ingested data.
Sanitization: No explicit sanitization or instruction-filtering is applied to the diff output before it is passed to the reviewers.

check