Core Concepts
Templates
A template is a versioned, installable bundle of agent components — skills, MCP servers, commands, sub-agents, playbooks, instructions, and arbitrary files. template pack captures a working agent into a template; template onboard installs it elsewhere with one command.
The workflow
brainbase template pack # bundle the current agent
brainbase template publish # upload to the registry
brainbase template search "code review" # find templates
brainbase template info acme/reviewer # registry details for a template
brainbase template onboard acme/reviewer # install or refresh
brainbase template list # show installed templates
brainbase template remove acme/reviewer # uninstall
Components
Each component type is stored in the place its harness already reads from at runtime — the CLI doesn't impose its own layout. Templates target any supported harness, including kafka. See Harnesses for the per-harness install paths.
| Component | What it is |
|---|---|
| Skill | A SKILL.md folder the harness loads on demand. |
| MCP server | A Model Context Protocol server config (URL or stdio command). |
| Command | A slash command / prompt the harness exposes. |
| Sub-agent | A scoped agent definition (Claude Code / Kafka). |
| Playbook | A markdown playbook plus an auto-managed table in the instructions file. |
| Instruction | A block injected into CLAUDE.md / AGENTS.md / KAFKA.md between markers. |
| File | An arbitrary file or directory copied to a target path. |
Identifiers and versions
Templates are addressed as creator/slug and resolve to the latest published version unless you pin one. Installs are tracked per machine in ~/.brainbase/installed.json, so template list and template remove know exactly what landed where.
Templates vs the agent manifest
Use the agent manifest to manage one live agent in sync with the cloud. Use templates to package a reusable starting point others can install into their own folders.