The Agent Skills Directory

[COMMAND_EXECUTION]: Performs standard development operations such as builds, tests, and linting to verify the state of a task.
[PROMPT_INJECTION]: Ingests content from the local repository (code, PRDs, and images) to build context, which constitutes an indirect prompt injection surface. The skill mitigates this through architect-level verification and an optional cleanup pass. Ingestion points: Repository inspection tools (Read/Grep/Glob/Bash) and user-provided artifacts. Capability inventory: File system access, shell execution, and sub-agent delegation. Sanitization: Relies on an independent architect sub-agent review and a mandatory deslop pass.

ralph