菜单
Delegates tasks to Manus when local capabilities are insufficient. Use when tasks require real browser automation, web scraping, persistent compute, long-running autonomous execution, or multi-step workflows with deliverables.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Identifies missing capabilities that warrant new skills. Analyzes repeated friction patterns, failed tasks, and user workarounds to recommend skill creation. Use when discovering-skills finds nothing suitable.
Finds existing skills before building new ones. Searches local installations, official Anthropic repository, and community collections. Evaluates fit and recommends installation or adaptation.
Audits and optimizes context, memory, and skill use for projects. Reviews CLAUDE.md files, .claude/rules/, and skill configurations for efficiency. Use when context feels bloated, skills aren't triggering correctly, or memory needs organization.
Modifies existing skills for new contexts. Forks skills from repositories, adjusts for local conventions, and customizes behavior while preserving core functionality. Use when a skill is close but not quite right.
| name | connecting-agents |
| description | Delegates tasks to Manus when local capabilities are insufficient. Use when tasks require real browser automation, web scraping, persistent compute, long-running autonomous execution, or multi-step workflows with deliverables. |
| allowed-tools | WebFetch, Bash, Read, Write |
Extend capabilities by delegating to external agents.
Delegate when:
Handle locally when:
Complete AI agent with autonomous multi-step execution.
Capabilities: Browser automation, file system, research, content creation, data processing
Use for: Web scraping, form filling, complex research, deliverables (reports, presentations, websites)
API: RESTful. Docs
import requests
response = requests.post(
"https://api.manus.im/v1/tasks",
headers={"Authorization": f"Bearer {MANUS_API_KEY}"},
json={"task": "Research competitor pricing and create summary report"}
)
task_id = response.json()["task_id"]
# Poll for completion, then fetch results
Results from external agents should be: