Delegate complex tasks to Manus AI - web research, report generation, code building, data scraping, file management, multi-turn conversations, project instructions. Full CRUD, Files, Projects, and Webhook APIs.
Instalar com Codex ou Claude Copie este prompt, cole no Codex, Claude ou outro assistente e deixe que ele revise a página da skill e instale para você.
Um comando direto ignora o prompt de revisão. Verifique a origem antes de executá-lo.
O comando permanece em uma só linha. Role horizontalmente para revisá-lo antes de copiar.
Prefere uma cópia local? Baixe os arquivos disponíveis atualmente no SkillsMP.
Exibindo SKILL.md
SKILL.md
Instruções da origem · Visualização somente leitura
name
manus
version
3.0.0
description
Delegate complex tasks to Manus AI - web research, report generation, code building, data scraping, file management, multi-turn conversations, project instructions. Full CRUD, Files, Projects, and Webhook APIs.
Delegate complex tasks to Manus - an autonomous AI agent that browses the web, writes code, generates reports, scrapes data, builds mobile apps, creates financial models, and delivers complete work products.
API Configuration
Base URL:https://api.manus.im/v1
Authentication Header:API_KEY: <your-key>
Set via:
MANUS_API_KEY environment variable
Or skills.manus.apiKey in OpenClaw config
Agent Profiles and Pricing (Manus 1.6)
IMPORTANT: The manus-1.5, manus-1.5-lite, and manus-1.5-max profiles are DEPRECATED and will stop working. Always use the 1.6 profiles below.
Profile
Speed
Depth
Best for
Relative cost
manus-1.6
Standard
Full
Most tasks, file creation, reports
Standard (recommended default)
manus-1.6-lite
Fast
Light
Quick lookups, simple questions, summaries
~40% cheaper
manus-1.6-max
Slow
Deep
Complex multi-step research, thorough analysis (+19.2% satisfaction vs standard)
~2x standard
Default: Always use manus-1.6 unless the user explicitly asks for lite or max.
Wide Research: All Wide Research sub-agents automatically use manus-1.6-max for maximum depth.
Cost estimation guidance
Before creating a task, inform the user of the expected profile:
Built-in connectors - Gmail, Notion, Google Calendar, Slack, Similarweb
Task Modes
Mode
Description
When to use
agent
Full autonomous agent with tool use
File creation, web browsing, code writing (recommended default)
chat
Conversational back-and-forth
Simple Q&A, brainstorming
adaptive
Auto-selects best approach
When unsure which mode fits
Default: Always use agent mode for tasks that produce deliverables.
Task Templates
Use these templates to create well-structured tasks. Always include createShareableLink: true.
Research Template
For market research, competitive analysis, topic deep-dives.
curl -s -X POST "https://api.manus.im/v1/tasks" \
-H "API_KEY: $MANUS_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"prompt": "Research [TOPIC]. Find the top [N] [entities] by [criteria]. For each, provide: name, description, key metrics, recent news. Compile into a structured report with sections, tables, and citations. Include a summary with key takeaways.",
"agentProfile": "manus-1.6",
"taskMode": "agent",
"createShareableLink": true
}'
Example prompts:
"Research the top 10 AI startups in healthcare by funding raised in 2025. For each provide founding year, total funding, key product, and notable customers."
"Do a competitive analysis of project management tools: Asana, Monday, Linear, Jira, ClickUp. Compare pricing, features, integrations, and user reviews."
"Research the current state of autonomous vehicle regulations in the US, EU, and China. Summarize key differences and recent changes."
Report Generation Template
For creating formatted documents, presentations, executive summaries.
curl -s -X POST "https://api.manus.im/v1/tasks" \
-H "API_KEY: $MANUS_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"prompt": "Create a [FORMAT] about [TOPIC]. Include: [SECTIONS]. Use professional formatting with headers, bullet points, and data tables where appropriate. Export as [FILE_TYPE].",
"agentProfile": "manus-1.6",
"taskMode": "agent",
"createShareableLink": true
}'
Example prompts:
"Create a 10-slide presentation about our company's AI strategy for the board meeting. Include market overview, our position, roadmap, and investment ask."
"Write an executive summary report on the state of remote work in 2025. Include statistics, trends, and recommendations. Export as PDF."
"Generate a one-page brief comparing AWS, GCP, and Azure for a startup choosing a cloud provider."
Code Building Template
For generating code projects, scripts, prototypes.
curl -s -X POST "https://api.manus.im/v1/tasks" \
-H "API_KEY: $MANUS_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"prompt": "Build a [LANGUAGE/FRAMEWORK] [PROJECT_TYPE] that [FUNCTIONALITY]. Include: [REQUIREMENTS]. Follow best practices for [LANGUAGE]. Include a README with setup instructions.",
"agentProfile": "manus-1.6",
"taskMode": "agent",
"createShareableLink": true
}'
Example prompts:
"Build a Python Flask API that takes a URL and returns a summary of the page content using BeautifulSoup for scraping. Include error handling, rate limiting, and a Dockerfile."
"Create a React dashboard component that displays real-time metrics from a WebSocket connection. Include TypeScript types and unit tests."
"Write a bash script that audits an AWS account for security misconfigurations: open S3 buckets, overly permissive IAM policies, unencrypted volumes."
Data Scraping Template
For extracting structured data from websites.
curl -s -X POST "https://api.manus.im/v1/tasks" \
-H "API_KEY: $MANUS_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"prompt": "Scrape [WEBSITE/SOURCE] and extract [DATA_FIELDS] for [ENTITIES]. Output as [FORMAT: CSV/JSON/table]. Include [N] results. Handle pagination if needed.",
"agentProfile": "manus-1.6",
"taskMode": "agent",
"createShareableLink": true
}'
Example prompts:
"Scrape the top 50 products from Product Hunt this week. Extract: name, tagline, upvote count, maker name, and URL. Output as CSV."
"Go to news.ycombinator.com and collect the top 30 stories. Extract title, URL, points, and comment count. Output as JSON."
"Find the top 20 open source AI projects on GitHub by stars in the last month. Extract repo name, description, stars, language, and last commit date."
Allow Manus to ask follow-up questions (task pauses with stop_reason: ask)
projectId
string
-
Associate with a project's persistent instructions
taskId
string
-
Continue an existing conversation (multi-turn)
locale
string
-
Language locale (e.g., "zh-CN", "ja", "es")
attachments
array
-
Files to attach (see Attachment types below)
Attachment types:
// From an uploaded file ID (see Files API){"type":"fromFileId","fileId":"file_abc123"}// From a URL (Manus downloads it){"type":"fromUrl","url":"https://example.com/data.csv"}// Inline base64 content{"type":"fromBase64","data":"base64-encoded-content","fileName":"image.png","mimeType":"image/png"}
curl -s -X PUT "https://api.manus.im/v1/tasks/{task_id}" \
-H "API_KEY: $MANUS_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"title": "New title for this task"
}'
{"task_id":"abc123","status":"completed","task_title":"Research Report","output":[{"content":[{"type":"text","text":"Here is the research summary..."},{"type":"output_file","fileUrl":"https://private-us-east-1.manuscdn.com/...","fileName":"research_report.pdf"}]}]}
Always do these steps when a task completes:
Extract text content and present it to the user
List all output files with their names
Download files locally using curl
Deliver files to the user directly (do not rely on manus.im share links - they can be unreliable)
Download Output Files
# List files from a completed task
bash scripts/manus.sh files <task_id>
# Download all output files to a directory
bash scripts/manus.sh download <task_id> ./manus-output
Projects let you set persistent instructions that apply to all tasks in a project. Useful for brand guidelines, output format preferences, or domain context.
Create a Project
curl -s -X POST "https://api.manus.im/v1/projects" \
-H "API_KEY: $MANUS_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"name": "Q1 Market Research",
"instructions": "Always format output as markdown tables. Use formal tone. Include source URLs for all claims. Target audience: C-suite executives."
}'
Verify the agent profile is valid (manus-1.6, manus-1.6-lite, manus-1.6-max)
Verify the task mode is valid (agent, chat, adaptive)
Retry once after a 5-second wait
If it fails again, report the error to the user
Task failed during execution
When status is failed:
Report the failure to the user
Check the response for an error message or reason
Suggest simplifying the prompt or breaking it into smaller tasks
Offer to retry with a different agent profile
Rate limiting
HTTP 429
Wait 30 seconds before retrying
Inform the user: "Manus rate limit reached. Waiting before retry."
After 3 failed retries, stop and tell the user to try again later
Network or timeout errors
Retry the request once after 5 seconds
If polling times out (no completion after 10 minutes), inform the user the task is still running on Manus servers and give them the task ID to check later
Helper Script Reference
The included scripts/manus.sh provides CLI shortcuts:
manus.sh create "prompt" [profile] - Create a new task
manus.sh get <task_id> - Get full task details
manus.sh status <task_id> - Get status (pending/running/completed/failed)
manus.sh wait <task_id> [timeout] - Wait for completion (default: 600s)
manus.sh files <task_id> - List output files
manus.sh download <task_id> [dir] - Download all output files
manus.sh list [status] - List tasks (optionally filter by status)
manus.sh update <task_id> "new title" - Update task title
manus.sh delete <task_id> - Delete a task
manus.sh upload <filepath> - Upload a file, returns file_id
manus.sh list-files - List recent uploaded files
manus.sh delete-file <file_id> - Delete an uploaded file
manus.sh create-project "name" "instructions" - Create a project
manus.sh list-projects - List projects
manus.sh webhook-create <url> [events] - Create a webhook
manus.sh webhook-delete <webhook_id> - Delete a webhook
Need real-time web search? Use /parallel for fast multi-engine web searches that return immediately - good for gathering context before creating a Manus task.
Need social media / trend research? Use /last30days for Twitter/X, Reddit, and news monitoring with sentiment analysis.
Need to read a specific webpage? Use your built-in WebFetch tool rather than sending a full Manus task for a single page.
Manus is best for tasks that require autonomous multi-step work - browsing multiple sites, compiling findings, generating formatted deliverables. For quick lookups, other tools are faster and cheaper.
