The Agent Skills Directory

[SAFE]: The skill originates from a trusted organization (@anthropic-ai/claude-code).
[COMMAND_EXECUTION]: The skill orchestrates the execution of background tasks and shell commands (e.g., git, gh, test runners). These operations are appropriately scoped to the skill's purpose of planning and executing code refactors and are not used maliciously.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by delegating user instructions to background workers.
Ingestion points: The $instruction variable in SKILL.md captures raw user input.
Boundary markers: Absent. The worker instructions template does not use specific delimiters or instructions to treat the coordinated task as untrusted data.
Capability inventory: Worker agents possess the ability to modify the filesystem, run test scripts, and create pull requests on GitHub.
Sanitization: No sanitization is performed on the user instructions before propagation.
Note: This design is characteristic of orchestration skills and the risk is mitigated by the intended primary use case.

batch