Skills Architecture

What Are Skills

Camel Kit uses Anthropic Agent Skills to provide component-specific documentation and configuration guidance. Instead of loading all 396 Camel component catalogs into context, skills are loaded on-demand only when needed.

• Commands (like /camel-flow, /camel-implement) ask interactive questions about the integration
• Skills (one per Camel component, 396 total) provide component knowledge: description, capabilities, usage patterns, URI syntax, parameters, Maven dependencies, and reference to the complete JSON schema

Skills are user-invocable: false — they are only loaded by commands when specific components are selected.

Skill Structure

Each skill directory contains:

{agent-folder}/skills/camel-component-{name}/
├── SKILL.md          # Component documentation with YAML frontmatter
└── schema.json       # Complete component schema from Camel catalog

Where {agent-folder} is:

• .bob/skills/ for IBM Project Bob
• .gemini/skills/ for Gemini CLI
• .claude/skills/ for Claude Code

SKILL.md Format

---
name: camel-component-{name}
description: {component description}
user-invocable: false
---

# {Component Title}

{Component description}

## Component Information
- Scheme: {scheme}
- Syntax: {syntax}
- Type: Consumer/Producer/Both

## Usage Patterns
### Consumer (from)
- from:
    uri: {scheme}:name
    parameters:
      # See schema.json

## Configuration Reference
See [schema.json](schema.json) for complete options.

## Maven Dependency
<dependency>
  <groupId>org.apache.camel</groupId>
  <artifactId>{artifactId}</artifactId>
</dependency>

Pre-Generated Skills

All 396 Camel component skills are pre-generated and bundled in the camel-kit-core module:

Source location: camel-kit-core/src/main/resources/skills/

During camel-kit init, skills are copied to the appropriate AI agent folder:

• camel-kit init --ai bob copies to .bob/skills/
• camel-kit init --ai gemini copies to .gemini/skills/
• camel-kit init --ai claude copies to .claude/skills/

Why pre-generated?

1. Zero setup — Users get skills immediately during init
2. Version locked — Skills match the Camel version used by camel-kit
3. Offline ready — No need to download Camel catalog or clone repositories
4. Consistent — All users have identical component knowledge
5. Single source — One set of skills for all AI agents

Progressive Disclosure

Skills use a three-level progressive disclosure model:

Example flow:

1. User runs /camel-flow order-processing
2. Command asks questions, user mentions “Kafka” and “SQL”
3. Command loads ONLY the kafka and sql skills (2 of 396)
4. Context stays small, focused on relevant components

How Commands Use Skills

/camel-flow

When user selects a component:

1. Ask: “For [source], I suggest using: Component: kafka”
2. Reference: “For detailed documentation, see: .claude/skills/camel-component-kafka/SKILL.md”
3. Load skill: Read SKILL.md for usage patterns and capabilities

/camel-implement

When generating YAML:

1. Identify components from flow design
2. Load skills: .claude/skills/camel-component-{name}/SKILL.md
3. Read schemas: .claude/skills/camel-component-{name}/schema.json
4. Generate YAML with correct syntax and parameters
5. Validate against Camel YAML DSL schema

Examples

Kafka Component

Skill path: {agent-folder}/skills/camel-component-kafka/SKILL.md

name: camel-component-kafka
description: Send and receive messages to/from an Apache Kafka broker.
user-invocable: false

Usage Pattern:

- from:
    uri: kafka:topic
    parameters:
      brokers: "{{kafka.brokers}}"
      groupId: "{{kafka.groupId}}"

SQL Component

Skill path: {agent-folder}/skills/camel-component-sql/SKILL.md

name: camel-component-sql
description: Perform SQL queries using JDBC.
user-invocable: false

Usage Pattern:

- to:
    uri: "sql:INSERT INTO orders VALUES (:#id, :#amount)"
    parameters:
      dataSource: "#dataSource"

Extending: Adding a New Skill

1. Create a skill directory in camel-kit-core/src/main/resources/skills/:

skills/camel-component-{name}/
├── SKILL.md
└── schema.json

2. Follow the SKILL.md format above
3. Add MCP tool usage instructions if applicable
4. Register the skill in the appropriate command templates