Skip to content

Catalog Overview

The Bonsai catalog is the embedded collection of agents, abilities, and automation that ships inside the binary. Everything in the catalog is available to install into your project using the CLI.

Agent Types

6 agents — Backend Agent, DevOps Agent, Frontend Agent, Full-Stack Agent, Security Agent, Tech Lead Agent. Each has a specialized identity, defaults, and role.

Browse agent types

Skills

17 skills — Domain knowledge and standards. Coding conventions, API design, testing strategy, infrastructure patterns, and more.

Browse skills

Workflows

10 workflows — Step-by-step procedures for planning, code review, security audits, PR review, reporting, and implementation.

Browse workflows

Protocols

4 protocols — Hard rules that every agent must follow. Memory management, scope boundaries, security enforcement, and session startup.

Browse protocols

Sensors

13 sensors — Automated hook scripts that enforce boundaries, inject context, review output, and monitor code quality.

Browse sensors

Routines

8 routines — Periodic self-maintenance tasks. Backlog hygiene, dependency audits, doc freshness checks, vulnerability scans, and more.

Browse routines

From the CLI:

Terminal window
bonsai catalog

This displays a table of all available items grouped by type, with descriptions and compatible agent types.

From this documentation: Use the links above or the sidebar to navigate to any category page.

Items are installed during bonsai init (for the first agent) and bonsai add (for additional agents). Both commands present an interactive picker where you select which skills, workflows, sensors, and routines to install.

Terminal window
# Initialize project with Tech Lead — picks abilities interactively
bonsai init
# Add another agent — picks abilities for that agent type
bonsai add

Each agent type has defaults pre-selected in the picker. Required items (like protocols and core sensors) are auto-installed and cannot be unchecked.

The catalog is embedded into the Bonsai binary at build time from the catalog/ directory in the source repo:

catalog/
├── core/ # Shared core files (memory, self-awareness)
├── agents/ # Agent type definitions + identity templates
├── skills/ # Domain knowledge (meta.yaml + content.md)
├── workflows/ # Step-by-step procedures (meta.yaml + content.md)
├── protocols/ # Hard rules (meta.yaml + content.md)
├── sensors/ # Hook scripts (meta.yaml + script.sh.tmpl)
├── routines/ # Periodic tasks (meta.yaml + content.md.tmpl)
└── scaffolding/ # Project infrastructure templates

Every catalog item has a meta.yaml that defines its name, description, and compatible agent types. See the meta.yaml Schema reference for the full format.