tw-nlp/deepagentforce/web-search

Skill: Web Search (Tavily)

Search the public internet via Tavily API and return structured results for synthesis.

Applicable scenarios:

• Current events, breaking news, recent announcements
• Real-time data: weather, stock prices, sports scores, exchange rates
• Product research, technical documentation, third-party references
• Any fact that may be outdated in the model's training data

Execution

Command format (strictly follow — do not modify):

python <DEEPAGENTFORCE_ROOT>/src/services/skills/research/web-search/scripts/web_search.py "<搜索词>" [--max-results N] [--output file.json]

How to determine `<DEEPAGENTFORCE_ROOT>`:

<DEEPAGENTFORCE_ROOT> is the absolute path to the DeepAgentForce project root on the current host. Resolve it at runtime before executing:

find / -type d -name "DeepAgentForce" 2>/dev/null | head -1
# Example result: /home/user/projects/DeepAgentForce

Parameters:

Examples:

✅ Basic search:

python /home/user/projects/DeepAgentForce/src/services/skills/research/web-search/scripts/web_search.py "2024年诺贝尔奖得主"

✅ More results:

python /home/user/projects/DeepAgentForce/src/services/skills/research/web-search/scripts/web_search.py "latest React 19 features" --max-results 10

✅ Save to file:

python /home/user/projects/DeepAgentForce/src/services/skills/research/web-search/scripts/web_search.py "machine learning trends 2025" --output results.json

❌ Relative path (src/services/...) → execution will fail ❌ Missing quotes around query with spaces → argument parsing error ❌ Using this skill for internal HR/policy questions → use rag-query instead

Output Format

The script outputs structured JSON to stdout:

{
  "query": "搜索词",
  "total_results": 5,
  "results": [
    {
      "title": "页面标题",
      "url": "https://...",
      "snippet": "摘要内容（最多500字符）"
    }
  ]
}

Parse this output to synthesize a final answer for the user. Do not return raw JSON directly.