home / skills / dmmulroy / .dotfiles / websearch

websearch skill

safe

/home/.pi/agent/skills/websearch

This skill helps you fetch up-to-date facts and external docs by querying the web with the Exa-backed websearch tool.

npx playbooks add skill dmmulroy/.dotfiles --skill websearch

Review the files below or copy the command above to add this skill to your agents.

Files (1)

SKILL.md

1.1 KB

---
name: websearch
description: Search the public web using the Exa-backed websearch tool. Use this when the user asks for up-to-date facts, external docs, release notes, APIs, or anything not reliably present in the local repo.
---

# Websearch Skill

Use this skill whenever the task requires information from the internet.

## Tool to use

- `websearch`

## Workflow

1. Turn the user request into a focused query.
2. For time-sensitive requests, include the current year in the query.
3. Run `websearch` with sensible defaults:
   - `numResults`: `8`
   - `type`: `auto`
   - `livecrawl`: `fallback`
4. If results are weak, retry with:
   - broader or narrower wording
   - `type: "deep"` for harder research tasks
   - 1-2 follow-up searches instead of one huge query
5. Summarize findings clearly and include source URLs when available in results.

## Notes

- Prefer multiple targeted searches over one vague search.
- If a search fails or times out, explain that to the user and retry with a simpler query.
- If the request is purely local-repo work, do not use websearch.

Overview

This skill performs live web searches using an Exa-backed websearch tool to fetch up-to-date facts, external documentation, release notes, APIs, and other internet-native content. It is designed to be used whenever information cannot be reliably found in the local repository or requires current data. The skill returns concise summaries and source URLs from search results.

How this skill works

The skill converts the user's request into focused search queries, including the current year for time-sensitive topics. It runs the websearch tool with sensible defaults (numResults: 8, type: auto, livecrawl: fallback) and retries with adjusted queries or type: "deep" when results are weak. Findings are summarized clearly and include result URLs so users can verify sources.

When to use it

To fetch current facts, news, or release notes not present locally
When the user asks for external API docs, examples, or third-party guides
For research that needs multiple web sources or cross-checking
To verify claims, find authoritative sources, or link to primary docs
When local repository data is insufficient or potentially out of date

Best practices

Turn requests into narrow, focused queries; prefer multiple targeted searches over a single broad one
Include the current year for time-sensitive queries to avoid stale results
Start with default search settings and retry with broader, narrower, or deep searches if results are weak
Limit retries to 1–2 follow-ups and adjust wording rather than piling on many simultaneous variations
Always summarize key findings and include source URLs so users can inspect originals

Example use cases

Summarize the latest release notes for a third-party library and link to the changelog
Locate and extract relevant API endpoints and examples from external documentation
Compare recent news coverage or blog posts about a technology trend
Verify a claim by finding authoritative sources and quoting evidence with links
Find up-to-date configuration guidance or migration steps from vendor docs

FAQ

What if the search times out or returns weak results?

Explain the failure, retry with simpler or reworded queries, and use type: "deep" for tougher research tasks.

When should I avoid using this skill?

Avoid it for tasks that are purely local-repo work or when all needed data is already available locally.