research-web
Deep web research with parallel investigators, multi-wave exploration, and structured synthesis. Spawns multiple web-researcher agents to explore different facets of a topic simultaneously, launches additional waves when gaps are identified, then synthesizes findings. Use when asked to research, investigate, compare options, find best practices, or gather comprehensive information from the web.\n\nThoroughness: quick for factual lookups | medium for focused topics | thorough for comparisons/evaluations (waves continue while critical gaps remain) | very-thorough for comprehensive research (wave
Profile →npx skills add majiayu000/claude-skill-registry --skill research-web --agent claude-code
Same command for any agent — swap --agent for codex, cursor, copilot.
Weekly change comes from our own snapshots, not the repository page — it measures attention, not adoption.
What it does
Orchestrates parallel web researchers to comprehensively investigate a topic through iterative waves, then synthesize findings into actionable intelligence. Determines thoroughness level from user input or query characteristics, launches multiple web-researcher agents to explore distinct facets, and continues waves to address gaps. Uses an external orchestration file to track progress and synthesis.
How it works
- Determine thoroughness before researching using explicit hints or auto-selection rules. If the user specifies a level, honor it; otherwise auto-select based on query type (single fact → quick; focused topic → medium; comparison/evaluation → thorough; deep dive → very-thorough).
- Decompose the topic into orthogonal facets and assign 1–4 agents per wave depending on thoroughness level (medium: 1–2 agents; thorough: 2–4; very-thorough: 4–6).
- Launch Wave 1 with parallel agents using the task tool:
subagent_type: "vibe-workflow:web-researcher". Provide a broad exploration prompt and bound the scope for each facet. - After each wave, collect findings, cross-reference, classify gaps (Critical, Significant, Minor), and decide whether to launch subsequent waves according to satisficing criteria.
- Maintain an orchestration file at
/tmp/research-orchestration-{topic-slug}-{YYYYMMDD-HHMMSS}.mdfor external memory tracking, including Wave Tracking, Research Assignments, Collected Findings, Cross-Reference Analysis, Gap Evaluation, and Synthesis Notes. - Synthesize findings into final output after completing waves, ensuring all findings are sourced and integrated.
When to use it
- Use when asked to research, investigate, compare options, find best practices, or gather comprehensive information from the web.
- Automatically selects thoroughness when not explicitly specified, and continues waves while critical gaps or diminishing returns are detected.
What it can touch
- Launches web-researcher agents via the Task tool with
subagent_type: "vibe-workflow:web-researcher". - Writes to and reads from
/tmp/research-orchestration-{topic-slug}-{YYYYMMDD-HHMMSS}.mdas external memory for progress and synthesis.
Caveats
- License: MIT
- The process relies on multi-wave exploration and may continue until satisficing criteria are met or diminishing returns are observed; no hard maximum on waves is stated.
- Explicitly documents gaps and gating conditions to avoid unnecessary waves.
**Research request**: $ARGUMENTS # Thoroughness Level **FIRST**: Determine thoroughness before researching. Parse from natural language (e.g., "quick lookup", "thorough research", "comprehensive analysis") or auto-select based on query characteristics. **Auto-selection logic**: - Single fact/definition/date → quick - Focused question about one topic → medium - Comparison, evaluation, or "best" questions → thorough - "comprehensive"/"all options"/"complete analysis"/"deep dive" → very-thorough **Explicit user preference**: If user explicitly specifies a thoroughness level (e.g., "do a quick lookup", "thorough research on X"), honor that request regardless of other triggers in the query. **Trigger conflicts (auto-selection only)**: When auto-selecting and query contains triggers from multiple levels, use the highest level indicated (very-thorough > thorough > medium > quick). | Level | Agents/Wave | Wave Policy | Behavior | Triggers | |-------|-------------|-------------|----------|----------| | **quick** | 1 | Single wave | Single web-researcher, no orchestration file, direct answer | "what is", "when did", factual lookups, definitions | | **medium** | 1-2 | Single wave | Orchestrat
- Wave Continuation by Level
- Gap Classification
- Satisficing Evaluation
- Wave Planning
- Phase 1: Initial Setup (skip for quick)
- 1.1 Get timestamp & create todo list
- 1.2 Create orchestration file (skip for quick)
- Phase 2: Topic Decomposition & Agent Assignment
- 2.1 Decompose the research topic into ORTHOGONAL facets
- 2.2 Plan agent assignments with explicit boundaries
- 2.3 Expand todos for each research area
- 2.4 Update orchestration file
- Phase 3: Launch Parallel Researchers
- 3.1 Launch web-researcher agents
What does the research-web skill do?
Deep web research with parallel investigators, multi-wave exploration, and structured synthesis. Spawns multiple web-researcher agents to explore different facets of a topic simultaneously, launches additional waves when gaps are identified, then synthesizes findings. Use when asked to research, investigate, compare options, find best practices, or gather comprehensive information from the web.\n\nThoroughness: quick for factual lookups | medium for focused topics | thorough for comparisons/evaluations (waves continue while critical gaps remain) | very-thorough for comprehensive research (wave
How do I install it?
Run `npx skills add majiayu000/claude-skill-registry --skill research-web --agent claude-code` — it drops the skill into your project so the agent can pick it up. Swap the --agent value for codex, cursor or copilot if you use one of those.
Where does this skill come from?
From majiayu000/claude-skill-registry, a repository with 534 stars. We read it straight from the repository tree rather than a submitted listing, so what you see here is what is actually published.
Is a popular skill a good skill?
Not necessarily. Stars measure attention, not adoption — a repository can trend for a week and be abandoned. That is why we show the weekly change from our own snapshots next to the total, instead of a single flattering number.