connecting-agents

jxucoder 的头像作者 jxucoder

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.

📖 中文导读: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. 以下正文为该技能的原始 SKILL.md 内容(英文),上方安装命令可直接使用。

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

Connecting Agents

Extend capabilities by delegating to external agents.

When to Delegate

Delegate when:

  • Task requires capabilities Claude lacks (real browser, persistent compute, OS access)
  • External agent has specialized domain expertise
  • Long-running autonomous execution needed
  • User explicitly requests external agent

Handle locally when:

  • Claude can complete the task directly
  • Latency/cost of delegation exceeds benefit
  • Task requires conversational back-and-forth

Supported Agents

Manus

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

Delegation Process

  1. Identify capability gap
  2. Select appropriate external agent
  3. Formulate task description
  4. Execute via API/MCP
  5. Monitor for completion
  6. Integrate results into current context
  7. Report outcome to user

Integration

Results from external agents should be:

  • Summarized concisely
  • Validated if possible
  • Presented with source attribution
  • Flagged if uncertain or incomplete