Tools, skills, and extension data
Tools
Built-in tools (read, search, edit, bash, …) run through the agent loop under tools.approvalMode and related policy. Edit paths share hashline verification. Catalog: Tools reference.
Skills
Filesystem skill packages (e.g. SKILL.md trees) discovered under the active profile’s skill dir. A project-local skills directory is never scanned. Unreadable or malformed skill files surface load warnings. See Skills.
Plugins
veyyon plugin / marketplaces install bundles of skills, MCP, hooks, and related assets. See Plugins.
MCP
External tools via Model Context Protocol: client config in the profile’s mcp.json (there is no project scope). See MCP.
Hooks
TypeScript modules with pi.on(...), not JSON command tables. See Hooks.
Related
- Configuration: approval mode and settings paths
- Mechanisms