The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes git diff to identify modified code and make commands (make format-rs, make lint-rs) to format the codebase. These are standard operations for development tools working with local repositories.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted content (comments and docstrings from a git branch) which is then used to drive the agent's code-editing actions. While this constitutes an attack surface where comments could contain malicious instructions, the risk is inherent to the skill's purpose and is constrained by strict instructions to focus on concision and length limits.
Ingestion points: git diff origin/main...HEAD output (SKILL.md)
Boundary markers: None
Capability inventory: File system writes (editing comments), subprocess calls (git, make)
Sanitization: None

review-verbosity