Caps
Caps are composable agent primitives for behavior, knowledge, connections, and input templates.
Cap kinds
| Kind | Purpose |
|---|---|
psyche | Behavioral guidance included in instructions |
skill | Task guidance, optionally with supporting files |
service | An MCP connection definition and usage guidance |
prompt | Reusable input text with template parameters |
Skills and services are discovered through descriptions; their full guidance is loaded on demand. Prompts expand into input without starting a runnable. Psyches supply guidance across model calls while selected. A skill describes how to do work; an agic or flow provides a callable implementation of work. Tools perform operations and are selected separately.
Cap composition
An agic can select a psyche defined in the same program:
psyche concise:
Prefer short, direct answers.
agic explain(_) -> Text:
psyches = psyche/concise
Explain {{_}}.Define caps inline, store them locally, or attach sources through configuration
or with.
Scope and precedence
Root caps are shared under one Toolang root, home caps belong to one
agent, and here caps belong to the current source. For the same
kind/name, precedence is here, then home, then root.
skill/review identifies a cap; its source ref locates
the content. Tool and workspace access are configured separately.
See Extend agents with caps and File formats.