The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/relay.mjs script executes the agy CLI and git commands as its primary function. It uses Node.js spawn and execFileSync to interact with these tools using absolute paths and properly constructed argument arrays. This implementation follows security best practices for sub-process management and is consistent with the skill's stated purpose.
[INDIRECT_PROMPT_INJECTION]: The skill processes task instructions (briefs) and codebase content which are passed to the agy CLI for implementation. This pattern inherently possesses an attack surface for indirect prompt injection. The skill extensively mitigates this risk through its design and documentation, requiring a human or orchestrating agent to perform a detailed diff review, re-run project gates (tests, linters), and verify all changes before committing them to the repository.
Ingestion points: Instructions are read from files or stdin in scripts/relay.mjs and passed to the agy tool.
Boundary markers: Documentation in references/writing-the-brief.md recommends using XML-like structure (e.g., <task>, <verification_loop>) to provide clear boundaries for the implementer.
Capability inventory: The relay script can execute the agy CLI and git. The agy tool itself has the capability to modify the local filesystem and execute arbitrary project commands (such as test runners).
Sanitization: The skill relies on post-execution verification steps, as detailed in references/review-and-land.md, where the orchestrator must confirm that the diff matches the brief and that all security/linting gates pass.

agy-delegate