The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses shell commands like git diff, git log, and git rev-parse with user-supplied inputs to analyze changes. While expected for its purpose, this constitutes an execution surface.
[PROMPT_INJECTION]: Indirect Prompt Injection: The skill ingests untrusted content from the repository, including code diffs and documentation, which may contain adversarial instructions.
Ingestion points: Data is read from git output and local files such as CODING_STANDARDS.md, CONTRIBUTING.md, and spec files.
Boundary markers: There are no defined boundary markers to separate trusted instructions from untrusted repository content.
Capability inventory: The skill can execute local commands and instantiate new sub-agents.
Sanitization: Content from the repository is interpolated into agent prompts without validation or escaping.

code-review