tavily-search

name	tavily-search
description	Use Tavily web search to find current benchmarks, repositories, documentation, datasets, and non-paper resources that help solve a computer vision task.
allowed-tools	read_file run_python_inline
compatibility	Requires outbound HTTPS access and a TAVILY_API_KEY environment variable.

Tavily Search

Use this skill when paper metadata alone is not enough and you need current web results such as GitHub repositories, benchmark pages, implementation notes, or dataset documentation.

Workflow

1. Use focused queries with the exact CV task, modality, and constraints.
2. Prefer results that include repositories, docs, benchmark leaderboards, or dataset pages.
3. Save URLs and short notes into the workspace so later rounds can refer to the same sources.
4. Use Tavily sparingly; broad web search can add noise.

Python Pattern

import json
import os
import requests

api_key = os.environ["TAVILY_API_KEY"]
payload = {
    "api_key": api_key,
    "query": "translation only image registration github benchmark",
    "search_depth": "advanced",
    "max_results": 5,
    "include_answer": False,
    "include_raw_content": False,
}
response = requests.post(
    "https://api.tavily.com/search",
    json=payload,
    timeout=30,
)
response.raise_for_status()
print(json.dumps(response.json(), indent=2))

Practical Notes

• Good query patterns:
  • <task> github
  • <task> benchmark
  • <task> dataset
  • <task> review
• Use Tavily after a quick literature pass, not before.
• Prefer stable project pages over random blog posts.