Deep Research

Deep Research Skill

Overview

This skill provides a disciplined, bounded methodology for conducting high signal-to-noise web research. Load this skill BEFORE starting any content generation task to gather necessary evidence from multiple angles while strictly respecting the research budget and avoiding endless crawling.

Research Budget & Hard Limits

Research must strictly adhere to the following bounded budget to ensure timely delivery:

⚠️ Hard Ceiling: Under NO circumstances should any single task exceed 8 searches or 5 page fetches. Once the budget is reached, you MUST immediately stop researching and begin synthesizing and producing output.

Core Principles

1. Signal-to-Noise Over Raw Quantity: The quality of your output depends on selecting authoritative, high-density sources, NOT the raw count of web requests.
2. Law of Diminishing Returns: The first 3 to 5 well-targeted searches yield 80% of critical insights. Subsequent queries have rapidly diminishing returns. Stop when the core landscape and key figures are confirmed.
3. No Endless Rabbit Holes: Do NOT follow every secondary reference or tangential company mentioned in articles. Focus strictly on the user's primary prompt.
4. Snippet-First Evaluation: Evaluate search snippets first. Most facts (dates, specs, funding amounts, market rankings) are already clear in snippets. Only use fetch_page for deep analysis or conflicting claims.

When to Use This Skill

Always load this skill when:

Research Questions

• User asks "what is X", "explain X", "research X", "investigate X"
• User wants to understand a concept, technology, or topic in depth
• The question requires current, comprehensive information from multiple sources
• A single web search would be insufficient to answer properly

Content Generation (Pre-research)

• Creating presentations (PPT/slides)
• Creating frontend designs or UI mockups
• Writing articles, reports, or documentation
• Producing videos or multimedia content
• Any content that requires real-world information, examples, or current data

Research Methodology

Phase 1: Broad Exploration (1-2 Queries)

Start with 1-2 broad searches to map the landscape:

1. Initial Survey: Search for the main topic with temporal awareness (<current_date>)
2. Identify Core Dimensions: Identify 2-3 essential pillars (e.g., technical breakthrough, market status, commercial bottlenecks)

Phase 2: Targeted Deep Dive (2-3 Queries)

For each essential dimension, run 1 focused query:

1. Specific Queries: Use precise terms (e.g., "humanoid dexterous hand 2026 DOF tactile sensor")
2. Selective Reading: Use fetch_page ONLY on the 2-3 most authoritative or data-rich reports.

Phase 3: Diversity & Validation (1 Query if needed)

Resolve conflicting figures or verify industry challenges:

• Check concrete metrics (costs, yield, production volume, MTBF)
• Identify industry consensus vs. marketing hype

Phase 4: Synthesis Check

Before proceeding to content generation, verify:

• Have I answered the core user intent with concrete data and examples?
• Have I captured the main technical bottlenecks and commercial reality?
• Have I reached or approached my search budget (3-6 queries)?

Stopping Rule: If you have sufficient information to address the user's prompt, or if you have reached your search budget, STOP searching immediately and start synthesizing. Gaps in non-essential trivia should be acknowledged in the final report under "Recommended Next Inquiries" rather than delaying delivery with more searches.

Search Strategy Tips

Effective Query Patterns

Copy code
# Be specific with context
❌ "AI trends"
✅ "enterprise AI adoption trends 2026"

# Include authoritative source hints
"[topic] research paper"
"[topic] industry analysis"
"[topic] market report"

# Use temporal qualifiers — always use the ACTUAL current year from <current_date>
"[topic] 2026"
"[topic] latest"

Temporal Awareness

Always check <current_date> in your context before forming ANY search query.

When to Use fetch_page

Use fetch_page sparingly and selectively:

• ✅ The source is an authoritative in-depth technical whitepaper or official OEM report
• ✅ The snippet indicates vital data tables, architectural breakdown, or benchmarks needed for the report
• ❌ Do NOT fetch generic news aggregators, social forums, or pages where the snippet already answered your question

Quality Bar

Your research is sufficient when you can clearly answer:

1. What is the current status and latest breakthrough?
2. What are the key metrics and real-world deployment cases?
3. What are the core challenges or limitations preventing scale?
4. Who are the representative players and what is their approach?

Common Mistakes to Avoid

• ❌ Endless rabbit holes: Searching every secondary company, component, or tangent mentioned
• ❌ Over-fetching: Calling fetch_page on every single search result (wasting time and tokens)
• ❌ Perfectionist trap: Trying to crawl the entire internet before typing a single sentence
• ❌ Ignoring the budget: Continuing past 6-8 searches when diminishing returns have set in
• ❌ Superficial single search: Relying on 1 generic search without cross-checking key numbers

Output

After completing research (strictly within 3-6 queries):

1. Deliver the synthesis immediately: Produce the structured report, document, or answer.
2. Structured Evidence: Cite specific sources and data points gathered.
3. Open Questions: If certain niche details remain unverified due to budget limits, note them transparently in an "Open Questions / Next Steps" section.