The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes local command-line interfaces for Claude and Codex via the Bash tool. This is the core functionality and is used to trigger peer reviews and image generation with specific security flags (read-only or workspace-write).
[PROMPT_INJECTION]: The skill has an indirect prompt injection surface as it passes untrusted repository data and user input to the peer agent. Ingestion points: User questions, git diffs, and repository files (e.g., CLAUDE.md). Boundary markers: The skill does not use specific delimiters or protective instructions for ingested content. Capability inventory: Peer agents are provided with tools for filesystem inspection or writing to the workspace. Sanitization: No validation or escaping is applied to ingested data before it is sent to the peer agent.

confer