The Agent Skills Directory

[COMMAND_EXECUTION]: The skill's primary function is executed via scripts/relay.mjs, which invokes the agy CLI and git using Node.js child_process methods. This is consistent with the skill's purpose of delegating tasks and reviewing repository changes.
[PROMPT_INJECTION]: The skill incorporates a framework for processing external coding instructions ('briefs').
Ingestion points: The relay.mjs script accepts task descriptions from files or standard input.
Boundary markers: The skill recommends using XML-style tags to structure the brief, providing clear boundaries for the AI implementer.
Capability inventory: The delegated tool can modify the file system and interact with git, while the relay script monitors the process.
Sanitization: The skill relies on the implementer's internal safety logic and emphasizes a required manual review of all changes before they are committed.
[DATA_EXFILTRATION]: The relay script identifies a data exposure risk where the task brief content is passed as a command-line argument, making it visible to other users on a shared machine via the system's process list.
[SAFE]: The skill uses official tools from a trusted organization (Google) and follows security best practices by requiring human intervention to review and commit changes. No obfuscation or persistence mechanisms were found.

agy-delegate