manus-openclaw-bridge（manus OpenClaw bridge 文档）

Manus OpenClaw Bridge

Use this skill to turn OpenClaw into a Manus launcher and result collector.

Requirements

Runtime dependencies:

• OpenClaw
• Bash
• curl
• Python 3.9+
• Node.js 18+ (required for slides JSON -> PPTX conversion)

Required local configuration:

• ~/.config/manus-openclaw-bridge/manus.env
• MANUS_API_KEY must be supplied by each installer
• MANUS_API_BASE must be supplied explicitly by each installer and must use https://

Read REQUIREMENTS.md for a concise dependency and configuration summary. Read SECURITY.md for the download and credential safety model.

Quick start

1. Read references/setup.md if Manus API access is not configured yet.
2. Put the Manus API key and Manus API base URL in ~/.config/manus-openclaw-bridge/manus.env.
3. Submit a task with scripts/manus_submit.sh or scripts/manus_prompt.sh.
4. Poll/download results with scripts/manus_get_task.sh or scripts/manus_wait_and_collect.py.
5. If the result is a slides JSON bundle, convert it with scripts/manus_slides_json_to_pptx.mjs.
6. If the user is on Feishu, return downloaded files with the normal OpenClaw message/file send flow.

Workflow

1. Submit a new Manus task

Use scripts/manus_submit.sh "<prompt>" when the prompt is already clean.

Use scripts/manus_prompt.sh "Manus, generate an image of a rainy forest" when the message may contain a leading Manus trigger and mixed punctuation.

Defaults:

• MANUS_AGENT_PROFILE=manus-1.6
• MANUS_TASK_MODE=agent
• MANUS_API_BASE must be set explicitly to your Manus API base URL

2. Inspect an existing task

Use scripts/manus_get_task.sh <task_id> to fetch the raw task payload.

3. Wait for completion and collect files

Use python3 scripts/manus_wait_and_collect.py <task_id> [timeout_seconds].

This writes downloaded files to tmp/manus/ under the skill folder and prints a JSON summary with:

• task_id
• status
• task_title
• task_url
• files[]
• raw

Download safety rules:

• only https:// URLs are accepted
• only Manus-controlled allowlisted hosts are accepted
• redirect targets are validated again after connection open
• if Manus returns any other URL, the script refuses to fetch it and records a download_error

4. Handle slides/report outputs

If Manus returns a slides JSON artifact instead of a ready-made .pptx, convert it with:

node scripts/manus_slides_json_to_pptx.mjs <slides.json> <output.pptx>

Read references/slides.md when the task is slide-heavy.

5. Return results to the user

Prefer this pattern:

• If Manus returns downloadable files, send the files.
• If Manus returns images, send image files directly.
• If Manus returns only text/status, send a short summary plus the task URL when available.
• If Manus is still running, acknowledge quickly and poll with a bounded interval instead of a tight loop.

Feishu return pattern

When the current conversation is Feishu:

• Download files first with scripts/manus_wait_and_collect.py.
• Send each saved file path back through OpenClaw’s normal reply/file mechanism.
• For images, prefer image/file upload rather than pasting long CDN URLs.
• If there are multiple files, send the best result first, then the rest if useful.

Read references/feishu-return.md when wiring file return behavior.

OpenClaw integration notes

• Keep secrets out of the skill package. Never hardcode MANUS_API_KEY in scripts or docs.
• Store keys and endpoints in ~/.config/manus-openclaw-bridge/manus.env on each recipient machine.
• For direct-message style image requests, a simple trigger is enough: strip Manus, or manus: and forward the remainder.
• For group chats, define an explicit trigger rule to avoid accidental job launches.
• If returning files to Feishu/Telegram/etc., use the platform’s native send/upload tool after manus_wait_and_collect.py downloads the outputs.
• If the request is long-running, prefer acknowledging first and completing asynchronously.

When to read references

• Read references/setup.md for first-time installation and environment setup.
• Read references/chat-patterns.md when wiring the Manus flow into OpenClaw message handling rules.
• Read references/feishu-return.md when returning Manus images/files into Feishu.
• Read references/feishu-skill-template.md when another OpenClaw user wants a ready-made Feishu routing pattern.
• Read references/openclaw-integration.md for a full end-to-end integration pattern.
• Read references/slides.md when handling deck-generation outputs.