AgentSkillsCN

web-search

利用GPT-5.2结合网络搜索与高阶推理能力，从互联网上搜寻最新资讯。当您需要研究复杂查询、跨多个来源获取最新信息、了解近期新闻、查阅最新文档，或核实自模型训练截止日期以来可能已发生变化的事实时，优先选择此功能，而非单纯的webfetch。该功能会返回引用的权威来源。

SKILL.md
--- frontmatter
name: web-search
description: Search the web for up-to-date information using GPT-5.2 with web search and high reasoning. Prefer this over webfetch when you need to research complex queries, find current info across multiple sources, get recent news, up-to-date documentation, or verify facts that may have changed since your training cutoff. Returns cited sources.

What this skill does

Searches the web using a reasoning model with web search capabilities. Returns current, cited information from the internet rather than relying on potentially outdated training data.

When to use this

  • •When you need current/recent information (news, updates, releases, etc.)
  • •When verifying facts that may have changed since training cutoff
  • •When researching topics that require up-to-date sources
  • •When the user asks about recent events or developments
  • •When handling complex queries that require reasoning and multiple searches

How to use

Run the search script with your query as an argument.

Query format: Use full sentences, paragraphs, or lists of questions - not keyword searches. The model handles detailed, multi-part prompts well.

bash
uv run ~/.config/opencode/skill/web-search/search.py "your detailed query here"

Examples:

bash
uv run ~/.config/opencode/skill/web-search/search.py "What are the latest features in Python 3.13? I'm particularly interested in the new REPL improvements and any performance enhancements."
bash
uv run ~/.config/opencode/skill/web-search/search.py "I need to understand how Font Awesome licenses their icons. Can I use their Pro icons in an open source project? What are the restrictions for SVG downloads versus the web font?"

Important notes

  • •Slow: This typically takes 1-3 minutes due to web searching and reasoning. Set an appropriate timeout (180000ms+ recommended).
  • •Requires OPENAI_API_KEY: Must be set in environment.
  • •Cost: ~$0.10-0.20 per query (depends on response length and number of searches).

Output format

The script outputs:

  1. •Cost breakdown (tokens, searches, total cost)
  2. •The response text with inline citations
  3. •A list of all sources with titles and URLs