Agent Extensions

Agent Extensions are modular units of functionality that augment what agents can do. Rather than building every capability into the agent itself, Codebolt lets you compose agent behaviour from discrete, reusable pieces.

Extension types

Type	What it is	Where it lives
Capabilities	Reusable bundles of agent behaviour with typed inputs/outputs	.codebolt/capabilities/
Skills	Slash-command invocable capabilities from the marketplace	~/.codebolt/skills.json
Action Blocks	Lightweight code units that run as side executions parallel to an agent	.codebolt/actionblocks/
Executors	Runtime environments that run capabilities (Node, Python, shell, etc.)	.codebolt/capabilities/executors/
Installing MCP Servers	Add MCP-compliant tool servers from the marketplace or manually	.codebolt/mcp-servers.yaml
Managing MCP Servers	Start, stop, update, and monitor installed MCP servers	.codebolt/mcp-servers.yaml

How they relate

System event (file saved, git commit, agent completes)
        │
        ▼
     Hook  ──────────────────────────────────────────┐
        │ triggers                                    │
        ▼                                             ▼
     Agent  ──── invokes ────▶  Capability / Skill   Action Block
                                      │
                                      ▼
                                  Executor
                             (runs the capability)

Agents invoke capabilities and skills during their reasoning loop. Action blocks run in parallel as side executions. Executors provide the language runtime that capabilities need. For event-driven triggers that start agents automatically, see Automation.

Where to find them

All extension panels are available from the Agents dropdown in the bottom bar:

• Capabilities → Agents → Capabilities
• Action Blocks → Agents → Action Blocks