NextClaw Harness SDK

NextClaw provides two headless entry points. They share task, session, event, and result semantics, so scripts, CI jobs, and Node.js programs can choose an entry point without changing core behavior.

Entry point	Use it for	What it provides
In-process Harness	Node.js applications, services, and long-lived processes	Compose Agents, manage sessions and runs, and own lifecycle in process
nextclaw exec	Shell, CI, scheduled jobs, and pipelines	One task with clean stdout and text/JSON/JSONL output
Platform capabilities	Agent platforms that bring their own integrations	Observe events, extend ingress, and register tools, context, models, runtimes, and MCP

Which should I use?

• For one task and one result, use nextclaw exec or runNextclawTask().
• For multiple tasks in one process, event callbacks, or explicit session control, use the Harness.
• To resume a session, pass sessionId or --session explicitly. Without it, a new exec:<uuid> session is created.

Quick start

The CLI is included with NextClaw:

nextclaw exec "Summarize the current workspace"

Install the dedicated Harness entry package for Node.js applications:

pnpm add @nextclaw/harness

import { runNextclawTask } from '@nextclaw/harness';

const result = await runNextclawTask({ input: 'Summarize the current workspace' });
console.log(result.text);

Experimental boundary

The current public surface covers Harness lifecycle, Agent and session handles, live runs, the Kernel-owned event bus and ingress, and lifecycle-scoped contributions for tools, context, models, runtimes, and MCP. App Server, structured output, approval handlers, storage adapters, sandbox adapters, skills, channels, and apps are not public Harness APIs yet.

Continue with the Harness API, Platform capabilities, nextclaw exec, and Examples.