AGENT SKILL

http-fetch

暂无简介

License 许可
MIT
最近更新
2026-11-14
Stars
★ 9
Forks
6

02 资源文件 1

03 SKILL.md 正文

HTTP Fetch Provider

Status: experimental Availability: unavailable unless a declared HTTP provider is active

Capability

Safe public-network HTTP GET with SSRF protection and redirect following. Retrieves raw web resources (HTML, PDF, Office files, etc.).

Required Capabilities

Capability Required Notes
http.fetch yes Provided by the agent runtime or an installed provider
document.html.extract no Needed to extract article text from raw HTML

Input

  • url: A publicly accessible HTTP/HTTPS URL
  • expected_content_type (optional): Hint for downstream processing

Output

  • Raw response bytes
  • Fetch receipt containing final_url, content_type, content_length, status_code

Success Conditions

  • HTTP 200 response
  • Content-Type matches or is compatible with downstream extraction
  • No redirect loop or SSRF redirect to internal host

Partial Conditions

  • Non-200 status codes may still yield useful content (redirects, auth gates)
  • JS-rendered pages: content will be the raw HTML, not the rendered DOM
  • Anti-bot platforms: may receive a challenge page instead of real content

Security

  • Block redirects to internal IP ranges (RFC 1918, loopback, link-local)
  • Never forward credentials, cookies, or auth tokens
  • Respect robots.txt and rate-limit responses

Runtime Boundary

The wheel still carries the connector's internal network.py implementation for oks-connector compatibility, including its SSRF and redirect checks. It is not a public Agent-facing Python API. Agents should use the active runtime HTTP capability or another declared Provider instead of importing fetch_url() directly.

04 同仓库更多 Skills