web-research-synthesis

“This skill defines a clear, bounded pattern for web research and synthesis. It declares specific tool usage (web_fetch with defined parameters), specifies rate limits and conflict-resolution rules, and includes explicit caveats. No prompt injection, tool overreach, impersonation, secret exfiltration, or spawn abuse detected. Scope is well-defined and appropriately narrow.”

---
name: web-research-synthesis
description: >
  Research a topic using live web fetches, then synthesise findings into a
  structured report with source citations. Use when an agent needs to go
  beyond static knowledge and pull current information from the web before
  answering.
triggers:
  - user asks about something time-sensitive or after your training cutoff
  - task requires verifying current API status, pricing, or availability
  - claim needs corroboration from a live source
steps:
  - Identify 2-5 candidate URLs (search page, official docs, changelog, status page)
  - Fetch each with web_fetch (markdown mode, maxChars 8000)
  - Extract the key facts relevant to the task
  - Cross-reference: prefer primary sources (official docs, changelogs) over summaries
  - Write a synthesis section: headline finding → supporting evidence → caveats
  - List all sources with URL and access timestamp
notes:
  - Always check official docs/changelogs for API endpoints before assuming they exist
  - Rate limits apply: do not loop-fetch the same domain more than 3 times per task
  - If a fetch returns 404 or 410, record it as a negative finding (not just silence)
  - If findings contradict each other, surface the conflict rather than picking one
---

# Web Research & Synthesis Skill

## Purpose
Agents often work from static training knowledge. This skill provides a
repeatable pattern for fetching live information, resolving conflicts between
sources, and producing a structured synthesis with citations.

## When to Use
- User asks about an API, service, or product that may have changed
- You need to verify whether a feature, endpoint, or pricing still exists
- A task requires current data (live prices, status, changelog entries)

## Core Pattern

### 1. Identify Sources
Before fetching, list candidate URLs:
- Official docs (most authoritative for APIs)
- Changelog / release notes (catches deprecations)
- Status pages (catches outages)
- Community threads (catches real-world gotchas)

### 2. Fetch & Extract
Use `web_fetch` with `extractMode: markdown` and `maxChars: 8000`.
Pull only what is relevant to the task. Note any 404/410 responses.

### 3. Cross-Reference
- Primary sources beat secondary
- Recent dates beat old dates
- Explicit deprecation notices beat silence

### 4. Synthesise
Structure your output:
```
## Finding
One-sentence headline.

## Evidence
- Source 1 (URL, date): what it says
- Source 2 (URL, date): what it says

## Caveats
What you could not verify. What might change.

## Sources
- [Official Docs](url) — accessed YYYY-MM-DD
```

## Key Lesson
If an API endpoint returns 410 Gone, check the changelog for the replacement.
Do not assume silence means the endpoint still works.