/scrape

description: "Use when the user asks to research URLs or topics across the web, compare sources, or produce a sourced research report. Fans out up to 6 parallel researchers, compiles the findings, and spot-checks every source and claim."

/team scrape — Web Research Pipeline

> Called by team.md router when action is `scrape`. > Parallel web research across URLs or topics.

Role Selection Bias

| Layer | Preferred Roles | Count | |-------|----------------|-------| | L1 | domain-researcher (1 per URL/topic, max 6) | 2-6 | | L2 | technical-writer (compilation) | 1 | | L3 | domain-accuracy | 1 | | L4 | principal-architect or principal-pm | 1 |

Input Parsing

• URLs (starts with http/https) → one L1 agent per URL
• Quoted topics → one L1 agent per topic (uses WebSearch)
• Mixed → one agent per item
• Maximum 6 L1 agents

Pipeline Customization

Layer 1: Parallel Scraping

• **Mode:** `default` (web tools)
• URL agents: WebFetch → extract key info, code examples, API details
• Topic agents: WebSearch → WebFetch top 3-5 results → synthesize findings
• Each agent reports: key facts, code snippets, source URLs, confidence level

Layer 2: Compilation

technical-writer compiles all L1 findings into a unified research document:

• Deduplicated findings
• Cross-source validation
• Structured by topic/theme

**Toolkit integration:**

• No specific toolkit entries apply — agents use WebSearch/WebFetch directly
• Check team:toolkit.md only if outputting a formal document (e.g., research report)

Layer 3: Source Verification

domain-accuracy spot-checks:

• Are source URLs still valid?
• Do claims match what the sources actually say?
• Any contradictory information between sources?

Layer 4: Synthesis

Principal produces executive summary with ranked findings and recommended actions.

GSD Integration

N/A — Web scraping is project-context-independent.