The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides tools for executing arbitrary shell commands across multiple packages. The bash tool in pi-coding-agent, pi-mom, and pi-pods allows agents to run commands on the local host, in Docker containers, or on remote GPU pods via SSH. Users can also execute commands directly from the editor using ! or !! prefixes. \n- [EXTERNAL_DOWNLOADS]: The coding agent includes a package manager for extending capabilities. The pi install command downloads and installs extensions, skills, and themes from npm registries and git repositories (e.g., npm:@scope/pkg, git:github.com/user/repo). Documentation explicitly warns that installed packages run with full system access and should be reviewed before installation. \n- [REMOTE_CODE_EXECUTION]: The system architecture supports runtime execution of external code modules. Extensions are implemented as TypeScript modules and loaded dynamically at runtime using jiti. The pi-pods component executes setup and deployment scripts on remote servers via SSH, including installing system dependencies and Python environments. \n- [CREDENTIALS_UNSAFE]: The skill manages sensitive credentials for numerous LLM providers. API keys and OAuth tokens are stored locally in JSON files (e.g., ~/.pi/agent/auth.json) and retrieved from environment variables. The pi-ai package facilitates authentication for 20+ providers including Anthropic, OpenAI, Google, and AWS Bedrock.

pi-mono