benjaminjackson/exa-skills

Overview

This skill generates factual, structured answers by combining AI synthesis with web citations. It focuses on token-efficient output and predictable schemas so you can get parseable results with lower cost. Use it when you need verifiable answers or consistently structured data extracted from web sources.

How this skill works

The skill runs a web-aware AI answer command that searches sources, synthesizes findings, and returns a structured response according to your chosen output format or schema. You can request toon for compact human-readable output, JSON for programmatic consumption, or supply a JSON Schema to enforce a consistent object wrapper for parseable fields. It also supports custom system prompts to control tone and level of detail.

When to use it

• When you need a concise, cited answer to a factual question
• When you require structured data (tables, arrays, or objects) extracted from search results
• When you want token-cost efficiency for repeated queries
• When you plan to pipe results into jq or other tools for automation
• When you need consistent, validated output across many queries

Best practices

• Choose one output strategy and stick with it: toon for reading, JSON+jq for field extraction, or schemas+jq for strict structure
• Always wrap schema definitions in a root object (type: "object") to avoid validation errors
• Avoid the --text flag unless you need full source text to save tokens
• Do not mix toon output with jq — toon is YAML-like, jq expects JSON
• Break complex shell flows into sequential steps rather than nested command substitutions

Example use cases

• Generate a short, cited summary of a technology with a --output-format toon for quick review
• Produce a JSON object of company metadata using --output-schema and pipe to jq for ingest into a database
• Extract a ranked list of resources as an array schema and format for consumption by another script
• Run repeated Q&A with consistent schema to populate an FAQ or knowledge base
• Use a custom system prompt to tailor explanations for different audiences (e.g., developer vs. non-technical)