think

name	think
description	Zero-base deep research: structured investigation → synthesis → proposal with counter-arguments
argument-hint	<topic> [github.com/owner/repo]
user-invocable	true
disable-model-invocation	true
allowed-tools	["Read","Glob","Grep","Write","Agent","Bash(gh issue list:*)","Bash(gh api:*)","Bash(git log:*)","Bash(git diff:*)","TaskCreate","TaskUpdate","TaskList","TaskGet","AskUserQuestion","WebSearch","WebFetch","mcp__gemini-deepsearch__deep_search","mcp__perplexity__perplexity_research","mcp__perplexity__perplexity_search","mcp__perplexity__perplexity_ask","mcp__perplexity__perplexity_reason","mcp__playwright__browser_navigate","mcp__playwright__browser_snapshot","mcp__playwright__browser_take_screenshot","mcp__playwright__browser_close"]

/think — Zero-Base Deep Research & Proposal

Zero-base thinking: discard assumptions, collect facts, synthesize essentials, propose with counter-arguments.

This is a long-running skill. Use TaskCreate to track phases and persist progress across /compact.

Principles

1. No guessing: Every claim requires a source URL. Unknown = say "unknown"
2. Zero-base: Discard existing assumptions. Build conclusions bottom-up from facts
3. MECE: Design research axes with no gaps or overlaps
4. Pyramid principle: Conclusion → rationale → data
5. Generator-Critic: Always verify proposals with counter-arguments

Modes

Mode	Trigger	Purpose
Normal	/think <topic>	General deep research → analysis → proposal
Repo Analysis	/think <topic> github.com/owner/repo	Repository competitive analysis → roadmap proposal

---

Normal Mode

/think <topic>
  │
  Phase 1: Scoping + Deep Search (auto)
  │  ├─ Research design (MECE axes)
  │  └─ Gemini DeepSearch + Perplexity MCP parallel execution
  │
  Phase 2: Research (integration + supplemental)
  │  ├─ Cross-validate Gemini/Perplexity results
  │  ├─ deep-researcher for supplemental (SNS, local, latest)
  │  └─ source-verifier for all URL validation
  │
  Phase 3: Deep Dive (parallel analysis)
  │  ├─ case-analyzer × N (parallel)
  │  ├─ social-scanner (reputation research)
  │  └─ Additional Deep Search via MCP as needed
  │
  Phase 4: Synthesis (identify essentials) ← Main agent
  │
  Phase 5: Proposal (propose + counter-argument verification)
       └─ counter-argument agent

Phase 1: Scoping + Deep Search

Goal: Design research axes and auto-execute deep searches.

1. Decompose the topic into 3-5 MECE research axes
2. For each axis, execute in parallel:
   • mcp__gemini-deepsearch__deep_search (free, 250/day) — primary
   • mcp__perplexity__perplexity_research — top 1-3 axes only (cost: ~$0.4-1.3/call)
3. TaskCreate for each axis to track progress

Phase 2: Research Integration

Goal: Cross-validate and fill gaps.

1. Compare Gemini and Perplexity results for consistency
2. Launch deep-researcher agent for supplemental research:
   • SNS/community discussions
   • Japan-local information
   • Latest developments not in deep search results
3. Launch source-verifier agent to validate all URLs
4. Flag contradictions between sources

Phase 3: Deep Dive

Goal: Detailed analysis of key cases and social impact.

Launch in parallel:

• case-analyzer agents for each important case/example (up to 5)
• social-scanner agent for community sentiment and reception
• Additional targeted deep searches if gaps remain

Phase 4: Synthesis

Goal: Extract the essential insight from all research. Main agent executes this.

1. What is the structural pattern across all findings?
2. What is the root cause / driving force?
3. What does this mean for the topic's future?
4. Distill into 1-3 key insights with supporting evidence chains

Phase 5: Proposal

Goal: Actionable proposals with counter-argument verification.

1. Generate 2+ proposals based on Phase 4 insights
2. Each proposal includes:
   • Summary and expected impact
   • Pros / cons / risks
   • Implementation approach
   • Evidence chain (fact → inference → conclusion)
3. Launch counter-argument agent to stress-test each proposal
4. Revise proposals based on counter-arguments

---

Repo Analysis Mode

/think <topic> github.com/owner/repo
  │
  Phase 0: Repo Analysis (understand the repository)
  │  ├─ repo-analyzer: code, features, Issues/PRs
  │  ├─ Build feature map
  │  └─ ★ Confirm feature map with user
  │
  Phase 1: Scoping + Deep Search (competitor/prior art survey)
  │  ├─ Research design scoped to repository's domain
  │  └─ Gemini/Perplexity MCP auto-execution
  │
  Phase 2-3: Research + Deep Dive (competitor analysis)
  │
  Phase 4: Gap Analysis (identify differentials)
  │  ├─ Feature comparison matrix
  │  ├─ Positioning analysis (strengths/weaknesses/opportunities/threats)
  │  └─ Essence: why chosen / why not chosen
  │
  Phase 5: Roadmap Proposal
       ├─ Short/mid/long-term prioritized actions
       ├─ 2 roadmap variants
       └─ counter-argument verification

★ = User interaction point

Phase 0: Repo Analysis

1. Launch repo-analyzer agent with the GitHub URL
2. Build feature map from analysis results
3. Ask user to confirm/correct the feature map before proceeding

Phase 4 (Repo Mode): Gap Analysis

Instead of general synthesis, perform:

• Feature comparison matrix (this repo vs competitors)
• Positioning analysis with axes relevant to the domain
• Identify: "Why would someone choose this over alternatives?"
• Identify: "Why would someone NOT choose this?"

Phase 5 (Repo Mode): Roadmap Proposal

1. Generate 2 roadmap variants:
   • Variant A: Focus on deepening existing strengths
   • Variant B: Focus on addressing critical gaps
2. Each with short-term (1-3 months), mid-term (3-6 months), long-term (6-12 months)
3. Launch counter-argument agent
4. Present both variants with counter-argument results

---

Deep Search Usage

MCP tools for auto-execution (no manual user action needed):

MCP	Model	Cost	Usage
mcp__gemini-deepsearch__deep_search	Gemini Deep Research	Free (250/day)	Primary research
mcp__perplexity__perplexity_research	sonar-deep-research	~$0.4-1.3/call	Supplemental, top axes only

Cost management: Run Gemini first for all axes → Perplexity for top 1-3 axes only.

---

Sub-Agents

Agent	Role	File
repo-analyzer	Collect GitHub repo features, Issues/PRs, external reviews	agents/repo-analyzer.md
deep-researcher	Web/SNS search to supplement Deep Search results. Collector only	agents/deep-researcher.md
case-analyzer	Detailed analysis of individual cases (outcomes, success factors, reception)	agents/case-analyzer.md
social-scanner	X/Reddit/HackerNews/community sentiment research	agents/social-scanner.md
source-verifier	Validate all URLs exist + check claim consistency	agents/source-verifier.md
counter-argument	Challenge proposals: counter-arguments, logical leaps, risks	agents/counter-argument.md

---

Output

Save to workspace/{topic}/:

• repo-analysis.md — Repository analysis & feature map (repo mode only)
• research.md — Collected information with sources
• analysis.md — Deep dive analysis
• proposal.md — Final proposals / roadmap (the deliverable)

---

Quality Criteria

• All factual claims have source URLs
• 2+ proposals (comparable alternatives)
• Each proposal has pros / cons / risks
• source-verifier: all URLs validated
• counter-argument: all proposals stress-tested
• Logic chain (fact → inference → conclusion) explainable to a third party