Manus — Autonomous Agentic Task Executor (programmatic)

Status: ✅ Live-tested 2026-07-15 (create task → 200 {task_id,url}; status → completed; model=manus-1.6-agent). Key: MANUS_API_KEY (env.sh). API: https://api.manus.ai/v1 — Header: API_KEY: <key> (NOT Bearer).

What it is

Manus is a "hands-on" agent that executes multi-step tasks end-to-end (research, build, browse, produce files) on its own cloud infra — returns a task_id + a shareable task_url on manus.im.

When to use

• Offload a long autonomous multi-step task to run on Manus's side while OpenClaw stays free (e.g. "build a landing page", "research X and produce a report", "compile dataset").
• Complements our own sub-agents/Kilo: Manus runs fully hosted + has its own browser/file tools; good as a parallel external executor.

Usage

. /data/.openclaw/secrets/env.sh
python3 skills/manus/manus.py create "<task prompt>" [--mode <mode>]
python3 skills/manus/manus.py status "<task_id>"

• create → returns task_id, task_url (open on manus.im to watch live).
• status → returns status (completed/running), messages, produced artifacts.

Cost note

Each task consumes Manus credits — use for tasks worth offloading, not trivial pings.