The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes local system tools including the AO orchestrator, GitHub CLI (gh), and tmux to manage coding environments, git worktrees, and parallel agent sessions. It also guides the user through setup commands that modify platform configurations and restart local services.
[EXTERNAL_DOWNLOADS]: The skill facilitates the installation of external software dependencies through automated project setup hooks (postCreate) using package managers like pnpm and npm.
[PROMPT_INJECTION]: The orchestrator is designed to process untrusted external content such as GitHub issue descriptions and pull request reviews. This creates a surface for indirect prompt injection where malicious instructions embedded in these external sources could attempt to override agent instructions. Ingestion points: External data enters through tools like ao_issues and ao_review_check. Capability inventory: The skill can spawn autonomous coding agents, send instructions to active sessions, and execute repository setup scripts. Boundary markers: The instructions include an explicit safety control (Rule 3) requiring the agent to obtain human approval before performing significant actions like spawning agents or batch processing. Sanitization: No specific content filtering or sanitization of external data is defined in the instructions.

agent-orchestrator