Adding an Agent Definition

1. What a definition is

One file in src/modules/agents/definitions/, exporting one new Agent({...}).

• Capability needing only a path (guidelinesPath, commandsDir, skillsDir): plain string field.
• Capability needing more shape (mcp): own configuration object, built with a create{Name}Configuration factory.
• Reference: definitions/claude-code.ts.

2. Steps

1. Create src/modules/agents/definitions/{name}.ts, default-exporting one new Agent({...}). Why default export: write-typescript-code section 4.
2. Set name (snake_case, machine-readable) and displayName.
3. Set detectSystemPathUsing: closure returning a DetectionConfiguration for CLI detection on the system, usually a command check.
4. Set detectProjectPathUsing: closure returning a DetectionConfiguration for detecting the agent in the current project, usually files or paths checks.
5. Set only capabilities the agent supports: guidelinesPath, commandsDir, skillsDir (string paths), mcp (McpConfiguration from createMcpConfiguration, with configPath and entryMappers).
6. Register in src/modules/agents/registry.ts: one import, one entry in agentDefinitions.

No other source file should change. If a sync stage needs touching, a capability abstraction leaks agent-specific logic. Fix the capability configuration, not the stage.

Two files outside the source tree need updating. refactor-to-agenteq and add-agenteq-to-boost-project each hold an identical table of every agent's native guidelines path, skills/commands directory, command file extension, and MCP config path, key path, and entry shape. They run in other repos and cannot read definitions/{name}.ts. When adding an agent, or changing any of those fields, update the row in both tables.

3. Related Skills

• General TypeScript conventions: write-typescript-code
• Adding a sync capability: add-capability-stage
• Verifying definition output: e2e-test-agents
• Reverse-engineering tables: refactor-to-agenteq, add-agenteq-to-boost-project