# Skill: Web Research

This skill improves the quality and reliability of web research tasks.

## Search Query Construction

- Write specific, targeted queries — avoid vague phrases like "how does X work"
- Include the domain/technology in the query: `"Node.js SQLite WAL mode performance benchmark"`
- Add `"site:github.com"`, `"official docs"`, or `"site:developer.mozilla.org"` for authoritative sources
- Add the current year to queries about recent developments to avoid stale results
- If a first query returns noise, rephrase using synonyms or the canonical technical term

## Source Quality

| Tier | Examples | Use for |
|------|----------|---------|
| Primary | Official docs, GitHub repos, RFCs, research papers | Facts, specifications, APIs |
| Secondary | MDN, Stack Overflow, major tech publications | Tutorials, best practices |
| Tertiary | Personal blogs, Medium posts, forum answers | Context only — verify claims independently |

Never cite a source you have not fetched. Search snippets can be misleading.

## Fetching Pages

- Only fetch URLs from primary or secondary sources
- If `web_fetch` returns an error or a login wall, note it as inaccessible and try an alternative URL
- Long pages: focus on headings and first paragraphs — read deeper only if the heading is directly relevant
- PDF links often don't fetch well — note them as "PDF — not fetched" if needed

## Handling Contradictions

If two sources contradict each other:
1. Note both claims and their sources
2. Check which source is more authoritative (tier)
3. Check which is more recent
4. Report the discrepancy explicitly — do not silently pick one

## Avoiding Hallucination

- Never invent URLs — only cite pages you actually fetched
- Never paraphrase a source as if it said something it didn't
- If you cannot find reliable information, state: "I was unable to find a reliable source for this claim"