domains / research / exa-web-search / exa-search

exa-search

Skill card

combined

The three copies implement the same Exa search/extract capability but with materially different procedures and constraints. The K-Dense copy is a detailed, versioned toolkit with a specific academic-source strategy, a two-pass scholarly pattern, and a mandatory tracking header. The wanshuiyin copy is a workflow skill with a specific argument-parsing contract, a strict script-resolution chain, and a structured output table. The sickn33 copy is a high-level, safety-focused wrapper that defers to an external connector and adds explicit limitations about data privacy and evidence quality. No single copy contains the full set of steps and constraints found in the others, so crowning one would hide real alternatives.

Repositories

Other copies

Combined skill

---
name: exa-search
description: "Searches scientific and technical web content with Exa and extracts page or PDF text from URLs. Supports scholarly discovery and broad web search."
---

# Exa Web Search

## Setup
1. Ensure `exa-py` is installed (`pip install exa-py` or `uv pip install "exa-py>=2.23.0,<3"`).
2. Set the `EXA_API_KEY` environment variable.
3. If using the K-Dense scripts, ensure the `x-exa-integration` header is set to `k-dense-ai--scientific-agent-skills`.

## Workflow
1. **Parse Request**: Identify if the user wants a topic search, URL extraction, or similar-page discovery.
2. **Locate Tool**: 
   - If in the ARIS environment, resolve `exa_search.py` via the `ARIS_REPO` chain.
   - Otherwise, use the local `scripts/exa_search.py` or a configured Exa connector.
3. **Execute Search**:
   - For scientific/technical queries, use `--category publication` and `--include-domains` with a scholarly allowlist (e.g., arxiv.org, nature.com).
   - For broad web queries, use standard search with optional `--category` (news, company, etc.).
   - For URL extraction, use the `get-contents` or `web-extract` capability.
4. **Present Results**:
   - Format results as a structured table including Title, URL, Date, and Key Content.
   - For academic hits, include author names and publication year.
   - Verify primary sources and report uncertainties; do not treat snippets as established facts.

## Conflicts
- **K-Dense-AI/scientific-agent-skills**: Requires the `x-exa-integration` tracking header and a specific two-pass academic strategy.
- **wanshuiyin/Auto-claude-code-research-in-sleep**: Requires the ARIS script resolution chain and a specific table format.
- **sickn33/agentic-awesome-skills**: Defers to an external connector and emphasizes privacy/safety limitations.