deep-research
npx skills add https://github.com/thepexcel/agent-skills --skill deep-research
Agent 安装分布
Skill 文档
Deep Research
Enhanced research engine for topics where training data is outdated.
Quick Start
Standard Mode
CLASSIFY â LANDSCAPE SCAN â SCOPE â HYPOTHESIZE â PLAN â [PLAN PREVIEW*] â RETRIEVE
â GAP ANALYSIS â TRIANGULATE â SYNTHESIZE â RED TEAM â SELF-CRITIQUE â PACKAGE
*Deep+ tier only
LANDSCAPE SCAN (MANDATORY – Before Anything Else)
[Search for OVERVIEW first - NO known entity names in query!]
WebSearch: "[topic] landscape overview [current year]"
WebSearch: "top [topic] list [current year]"
WebSearch: "[topic] ecosystem players [current year]"
â WRONG: "DeepSeek Qwen performance 2025" (uses names you already know)
â
RIGHT: "China open source LLM models list 2025" (discovers what exists)
â Extract ALL entity names from results
â List: Discovered (new to you) vs Confirmed (you knew)
â THEN proceed to SCOPE with complete picture
Why: You cannot research what you don’t know exists. Scan the landscape FIRST.
Creative Mode
ABSTRACT â MAP (3-5 domains) â SEARCH â GENERALIZE â SYNTHESIZE
Trigger: “creative mode”, “cross-industry”, “what do others do”
Example: “à¸à¸³à¸¢à¸±à¸à¹à¸à¹à¸«à¹à¸à¸à¸¡à¸² engage à¸à¸±à¸ online course มาà¸à¸à¸¶à¹à¸?” â ABSTRACT: “retention + engagement à¹à¸à¸à¸´à¸à¸à¸£à¸£à¸¡à¸à¸µà¹à¸à¸³à¸à¹à¸³” â MAP: Gaming (streaks, XP), Fitness apps (habit loops), YouTube (thumbnails, hooks), Loyalty programs (tiers) â SEARCH each domain â GENERALIZE patterns â SYNTHESIZE recommendations
Classification
| Type | When | Process | Example |
|---|---|---|---|
| A | Single fact | WebSearch â Answer | “Python 3.13 release date à¸à¸·à¸à¹à¸¡à¸·à¹à¸à¹à¸«à¸£à¹?” |
| B | Multi-fact | Scan â Retrieve â Synthesize | “à¹à¸à¸£à¸µà¸¢à¸à¹à¸à¸µà¸¢à¸ pricing à¸à¸à¸ cloud GPU providers” |
| C | Judgment needed | Full 6 phases | “à¸à¸§à¸£à¹à¸à¹ Next.js หรืภAstro สำหรัภblog?” |
| D | Novel/conflicting | Full + Red Team | “AI à¸à¸°à¹à¸à¸à¸à¸µà¹ data analyst ภายà¹à¸ 3 à¸à¸µà¸à¸£à¸´à¸à¹à¸«à¸¡?” |
Intensity Tiers
| Tier | Sources | When |
|---|---|---|
| Quick | 5-10 | Simple question |
| Standard | 10-20 | Multi-faceted |
| Deep | 20-30 | Novel, high stakes |
| Exhaustive | 30+ | Critical decision |
Search & Evidence
Parallel Search (MANDATORY)
[Single message â always 2-3 queries at once]
WebSearch: "[topic] [current year]"
WebSearch: "[topic] limitations"
WebSearch: "[topic] vs alternatives"
Claim Types
| Type | Requirements | Example |
|---|---|---|
| C1 (Key claim) | Quote + 2+ sources + confidence | “Next.js มี market share 42%” |
| C2 (Supporting) | Citation required | “Vercel à¹à¸à¹à¸à¸à¸¹à¹à¸à¸±à¸à¸à¸² Next.js” |
| C3 (Common knowledge) | Cite if contested | “React à¹à¸à¹à¸ library ยà¸à¸à¸à¸´à¸¢à¸¡” |
Confidence Format (C1 claims)
**Claim:** [Statement]
**Confidence:** HIGH/MEDIUM/LOW
**Reason:** [Why this confidence level]
**Sources:** [1][2]
Anti-Hallucination
- Every C1 cites [N] immediately
- Use “According to [1]…”
- Admit: “No sources found for X”
Research Sufficiency
“à¹à¸¡à¸·à¹à¸à¹à¸«à¸£à¹à¸à¸¶à¸à¸à¸°à¸à¸?”
| Signal | หมายà¸à¸§à¸²à¸¡à¸§à¹à¸² |
|---|---|
| Saturation | 3 sources à¸à¹à¸à¹à¸à¸·à¹à¸à¸à¹à¸¡à¹à¹à¸«à¹à¸à¹à¸à¸¡à¸¹à¸¥à¹à¸«à¸¡à¹ â à¸à¸à¹à¸¥à¹à¸§ |
| Convergence | หลาย sources สรุà¸à¹à¸«à¸¡à¸·à¸à¸à¸à¸±à¸ â confidence สูภ|
| Contradiction | Sources à¸à¸±à¸à¹à¸¢à¹à¸à¸à¸±à¸ â à¸à¹à¸à¸ dig deeper หรืภflag uncertainty |
| Diminishing returns | à¹à¸à¸´à¹à¸¡ search à¹à¸à¹à¹à¸à¹à¹à¸à¹ rephrase à¸à¸à¸à¹à¸à¸´à¸¡ â หยุà¸à¹à¸à¹ |
Quick tier: หยุà¸à¹à¸¡à¸·à¹à¸ saturation Standard: หยุà¸à¹à¸¡à¸·à¹à¸ convergence + gap analysis à¹à¸¡à¹à¹à¸à¸ gap สำà¸à¸±à¸ Deep/Exhaustive: หยุà¸à¹à¸¡à¸·à¹à¸ Red Team challenge à¹à¸¡à¹à¸à¸à¸à¸¸à¸à¸à¹à¸à¸à¹à¸«à¸¡à¹
Facilitation Guide
Progress Reporting
à¸à¸¸à¸à¹ 5-8 sources â update à¸à¸¹à¹à¹à¸à¹:
"สรุà¸à¸à¸µà¹à¸à¸à¸à¸à¸à¸¶à¸à¸à¸à¸à¸à¸µà¹: [key findings]
ยัà¸à¸¡à¸µà¸à¸³à¸à¸²à¸¡à¸à¹à¸²à¸: [gaps]
à¸à¸° search à¸à¹à¸à¹à¸£à¸·à¹à¸à¸ [next direction] à¸à¸°à¸à¸°"
When to Ask User
| สà¸à¸²à¸à¸à¸²à¸£à¸à¹ | à¸à¸²à¸¡à¸§à¹à¸² |
|---|---|
| Topic à¸à¸§à¹à¸²à¸à¹à¸à¸´à¸à¹à¸ | “à¸à¸¢à¸²à¸à¹à¸à¹à¸à¸¡à¸¸à¸¡à¹à¸«à¸à¸à¸°? [option A] หรืภ[option B]?” |
| à¹à¸à¸ sub-topic à¸à¹à¸²à¸ªà¸à¹à¸ | “à¹à¸à¸à¹à¸£à¸·à¹à¸à¸ X à¸à¸µà¹à¹à¸à¸µà¹à¸¢à¸§à¸à¹à¸à¸ â à¸à¸¢à¸²à¸à¹à¸«à¹à¸à¸¸à¸à¸¥à¸¶à¸à¹à¸«à¸¡à¸à¸°?” |
| Sources à¸à¸±à¸à¹à¸¢à¹à¸ | “à¹à¸«à¸¥à¹à¸ A à¸à¸à¸à¸§à¹à¸² X à¹à¸à¹à¹à¸«à¸¥à¹à¸ B à¸à¸à¸à¸§à¹à¸² Y â à¸à¸µà¹à¸£à¸° lean à¸à¸²à¸à¹à¸«à¸à¸à¸°?” |
| Deep+ tier, plan ready | “à¸à¸µà¹à¸à¸·à¸ plan สำหรัภresearch â approve à¸à¹à¸à¸à¹à¸à¸à¹à¸à¸à¸°à¸à¸°” |
Don’t Ask â Just Do
- Type A questions â à¸à¸à¸à¹à¸¥à¸¢
- Choosing search queries â à¸à¸³à¹à¸¥à¸¢ à¹à¸¡à¹à¸à¹à¸à¸à¸à¸²à¸¡
- Formatting output â à¹à¸à¹ template à¹à¸à¹à¹à¸¥à¸¢
Tools & Fallbacks
URL Fallback
If WebFetch returns 403:
curl -s --max-time 60 "https://r.jina.ai/https://example.com"
GitHub Repository Research
à¹à¸à¸ repo à¸à¹à¸²à¸ªà¸à¹à¸ â à¸à¸²à¸¡ user à¸à¹à¸à¸ clone:
"à¹à¸à¸ repo à¸à¸µà¹à¸à¹à¸²à¸ªà¸à¹à¸: [repo-name] â à¸à¹à¸à¸à¸à¸²à¸£à¹à¸«à¹ clone มาศึà¸à¸©à¸² code à¹à¸«à¸¡à¸à¸°?"
If agreed:
mkdir -p /mnt/d/githubresearch && cd /mnt/d/githubresearch && git clone [repo-url]
Key files: package.json/pyproject.toml â src/ main logic â README.md
References
| Topic | File | Grep Pattern |
|---|---|---|
| Phase details | standard-mode.md | grep -n "^## Phase" |
| Creative mode | creative-mode.md | grep -n "^## Phase C" |
| Agent prompts | agent-templates.md | grep -n "^## " |
| Progress/recovery | progress-recovery.md | â |
| Report template | report_template.md | â |
| Query generation | query-framework.md | QUEST Matrix |
| Perspective audit | perspective-checklist.md | COMPASS Checklist |
| Researcher thinking | researcher-thinking.md | THINK Protocol |
| Script | Purpose |
|---|---|
scripts/validate_report.py |
9-check quality validation |
Output File (MANDATORY)
After completing research, ALWAYS save to markdown file:
research/[topic-slug]-[YYYY-MM-DD].md
Example: research/china-opensource-ai-2025-01-04.md
- Create
research/folder if it doesn’t exist - Why: Research takes effort. Save it for future reference.
Related Skills
/boost-intelâ Apply critical thinking to research findings/generate-creative-ideasâ Creative Mode for cross-industry innovation/skill-creator-thepexcelâ Research domain expertise for skill creation/extract-expertiseâ Research to prepare expert interviews