analyze（分析）

Universal Analyzer

A standalone skill for multi-perspective analysis of any topic, file, idea, or decision.

Use when:

• Need to extract KEY POINTS from anything
• Want multi-perspective gap/risk analysis (4-10 experts)
• Need improvement opportunities with actionable plans
• Want quick summary OR deep analysis mode
• Need first-principles brief that challenges assumptions

Triggers: "analyze", "key points", "what's important", "improve this", "review", "examine", "assess", "analysis", "deep analysis", "run deep analysis", "brief", "challenge assumptions", "first principles"

Agent Capabilities

This skill remains fully functional without CLI tools. Codebase intelligence is an optional enhancement for the software domain only.

Table of Contents

1. Quick Start
2. Analysis Modes
3. Execution Flow
4. Output Formats
5. Perspectives Library
6. Thinking Framework
7. First-Principles Analysis
8. Examples

Quick Start

# Quick analysis (fast key points)
analyze quick "our pricing strategy"
analyze brief src/api/auth.ts

# Standard analysis (default)
analyze "SaaS product for developers"
analyze "should I accept this job offer"
analyze package.json

# Deep analysis (comprehensive)
analyze deep "company rebrand strategy"
analyze thorough "migration to microservices"

Analysis Modes

Execution Flow

Phase 0: Detection

1. Mode: Detect from keywords (quick/standard/deep)
2. Domain: Auto-detect from content/keywords
3. Target: Load file/folder content if path provided
4. TypeScript detection: If target is a code path and tsconfig.json exists:

- Detection gate: tsconfig.json exists + npx codebase-intelligence --help succeeds - For full CLI reference, fetch https://raw.githubusercontent.com/bntvllnt/codebase-intelligence/main/llms.txt - If available: gather structural data in Phase 1 - If unavailable: proceed with pure file-reading context gathering

Phase 1: Context Gathering

• File/folder path → Read files
• Software domain + TypeScript + CI available → Augment file reading with:

- npx codebase-intelligence overview --json <path> → project structure snapshot - npx codebase-intelligence modules --json <path> → module boundaries and cross-deps - npx codebase-intelligence groups --json <path> → directory-level aggregates Include as structured context alongside file contents. If CI unavailable: read files only.

• Topic/idea → Use provided context
• Unfamiliar domain → Optional web search

Phase 2: Analysis

Quick Mode

Direct analysis by primary agent, no sub-agents

Standard Mode

1. Select 4-6 perspectives based on domain
2. Launch all perspectives in parallel (single message block)
3. Each agent answers 7 core questions

Deep Mode

1. Select 6-10 perspectives based on domain
2. Launch all perspectives in parallel (single message block)
3. Each agent answers 12 questions (7 core + 5 deep)

Software Domain Enhancement (TypeScript + CI available)

Include pre-computed structural data in software domain agent prompts:

• npx codebase-intelligence hotspots --json --metric complexity --limit 20 → complexity/churn hotspots
• npx codebase-intelligence forces --json → cohesion/coupling tensions per module
• npx codebase-intelligence dead-exports --json --limit 20 → unused export analysis
• npx codebase-intelligence clusters --json → dependency community clusters

Agents receive this as structured context alongside the target content. Supplements their analysis — does not replace independent reasoning. If CI unavailable: agents analyze from file contents only.

Phase 3: Synthesis

After ALL agents complete:

1. Aggregate Failure Hypotheses → prioritize Critical→High→Med→Low
2. Extract TOP 10 key points → rank by consensus
3. Identify gaps → Critical/High/Medium/Low
4. Generate actionable roadmap with dependencies

Output Formats

Quick Mode

# Quick Analysis: [Target]

**Mode**: Quick | **Domain**: [Domain] | **Date**: [Date]

## The Essence
[1-2 sentences: What this ACTUALLY is at its core, stripped of complexity]

## Verified Facts
- [Fact 1] — evidence: `path/file:line` or [source]
- [Fact 2] — evidence: ...
- [Fact 3] — evidence: ...

## Key Points (Top 5-10)
1. **[Most Important]**: [Explanation]
2. **[Second]**: [Explanation]
...

## Assumptions to Challenge

| Assumption | Evidence For | Evidence Against | Verdict |
|------------|-------------|------------------|---------|
| [Assumed thing] | [If any] | [If any] | Validate/Keep/Discard |

## What You Haven't Considered

1. **[Critical Item]**: [Why this matters, what to do about it]
2. **[Hidden Risk/Opportunity]**: [Explanation]
3. **[Simpler Alternative?]**: [If exists, describe it]

## The Real Question
[Reframe what the user should actually be asking about this]

## Quick Actions (Top 3-5)
- [ ] [Action 1] - [Why/Impact]
- [ ] [Action 2] - [Why/Impact]
- [ ] [Action 3] - [Why/Impact]

## Critical Risk to Watch
[The one thing that could derail this]

Standard Mode

# Analysis: [Target]

**Mode**: Standard | **Domain**: [Domain] | **Perspectives**: [Count]
**Date**: [Date] | **Hypotheses**: [Count] total

## The Essence
[1-2 sentences: What this ACTUALLY is at its core, stripped of complexity]

## Executive Summary
[2-3 sentences capturing most critical findings]

## First-Principles Analysis

### Verified Facts
- [Fact 1] — evidence: `path/file:line` or [source] — confidence: High/Medium/Low
- [Fact 2] — evidence: ... — confidence: ...

### Assumptions to Challenge

| Assumption | Evidence For | Evidence Against | Verdict |
|------------|-------------|------------------|---------|
| [Assumed thing] | [If any] | [If any] | Validate/Keep/Discard |

### What You Haven't Considered

1. **[Critical Item]**: [Why this matters, what to do about it]
2. **[Hidden Risk/Opportunity]**: [Explanation]
3. **[Simpler Alternative?]**: [If exists, describe it]
4. **[Downstream Impact]**: [What this affects that wasn't mentioned]

### The Real Question
[Reframe what the user should actually be asking about this]

## Failure Hypotheses (Aggregated)

| ID | Type | IF (Trigger) | THEN (Failure) | BECAUSE | Sev | Mitigation |
|----|------|--------------|----------------|---------|-----|------------|
| S001 | Security | ... | ... | ... | Crit | ... |
| M001 | Misuse | ... | ... | ... | High | ... |

### Critical Mitigations Required
- [ ] {mitigation} ← Addresses: S001, M001

## Key Points (Ranked by Importance)

### Consensus Points (Flagged by 3+ perspectives)
1. **[Point]** - [Why important] (Flagged by: [Perspectives])

### Important Points
2. **[Point]** - [Explanation]

### Divergent Views
- **[Topic]**: [Perspective A] sees X, [Perspective B] sees Y
  - *Implication*: [What this tension means]

## Gap Analysis

### Critical Gaps (Action Required)
| Gap | Flagged By | Impact | Suggested Action |
|-----|------------|--------|------------------|
| ... | ... | High | ... |

### High Priority Gaps
[Grouped by theme]

## Improvement Opportunities

### Top 5 Improvements
1. **[Improvement]**: [Details] - Expected Impact: [High/Medium/Low]

## Action Plan & Roadmap

### Immediate Actions (This Week)
| Action | Owner | Dependencies | Success Criteria |
|--------|-------|--------------|------------------|
| ... | [TBD] | None | ... |

### Short-term (1-2 Weeks)
| Action | Dependencies | Resources Needed | Outcome |
|--------|--------------|------------------|---------|
| ... | Immediate #1 | ... | ... |

## Cross-cutting Concerns
[Issues flagged by 3+ perspectives]

Deep Mode

Deep mode extends Standard format with:

• Additional 5 questions per agent (12 total)
• Medium-term (1 Month) and Long-term (3+ Months) roadmap sections
• Dependency map visualization
• Risk mitigation table
• More detailed failure hypotheses

Perspectives Library

Domain Detection

Analyze content/keywords to auto-detect domain:

Perspective Roles by Domain

Business Domain

1. Startup Founder - Viability, growth, scalability
2. Investor - ROI, risk, market size
3. CFO - Unit economics, burn rate, margins
4. Customer - Value proposition, willingness to pay
5. Competitor - Differentiation, competitive moats
6. Market Analyst - TAM, trends, timing

Product Domain

1. Product Manager - User needs, roadmap, prioritization
2. Designer (UX) - Usability, user flows, accessibility
3. Customer - Actual use cases, pain points
4. QA Engineer - Edge cases, error states
5. Data Analyst - Metrics, success criteria
6. Support Lead - Maintenance burden, user confusion

Software Domain

1. Security Engineer - Vulnerabilities, attack vectors
2. DevOps - Deployment, monitoring, reliability
3. Architect - Design patterns, scalability, debt
4. QA Engineer - Test coverage, edge cases
5. Performance Engineer - Bottlenecks, resource usage
6. Tech Lead - Maintainability, team velocity

Process Domain

1. Operator - Daily execution, friction points
2. Manager - Efficiency, bottlenecks, metrics
3. Employee - Experience, clarity, pain points
4. Auditor - Compliance, documentation, risks
5. Improvement Specialist - Waste, optimization

Personal Domain

1. Future You (1 year) - Long-term impact
2. Future You (5 years) - Career trajectory
3. Skeptic - Risks, downsides, failure modes
4. Supporter - Strengths, opportunities
5. Financial Advisor - Money implications
6. Life Coach - Values alignment, fulfillment

General Domain (Fallback)

1. Analyst - Facts, data, patterns
2. Critic - Weaknesses, risks, gaps
3. Advocate - Strengths, opportunities
4. Pragmatist - Feasibility, resources, timeline
5. Strategist - Long-term implications, alternatives
6. Devil's Advocate - Unconsidered downsides

Per-Agent Prompt Template

Standard Mode (7 questions):

You are a [Role]. Analyze:
TARGET: [content]

Answer these 7 questions:
1. KEY POINTS: 3-5 most important elements?
2. CORE INSIGHT: Single most critical thing?
3. GAPS: What's missing or incomplete?
4. RISKS: What could go wrong?
5. ASSUMPTIONS: What needs validation?
6. IMPROVEMENTS: Top 3 ways to improve?
7. BLIND SPOTS: What isn't being considered?

FAILURE HYPOTHESES: For each risk/gap:
| ID | IF (Trigger) | THEN (Failure) | BECAUSE | Severity | Mitigation |

Include MISUSE + ADVERSARIAL hypotheses.
Rate findings: Critical/High/Medium/Low

Deep Mode (12 questions = 7 core + 5 additional):

1. DEPENDENCIES: What does success depend on?
2. ALTERNATIVES: What other approaches should be considered?
3. TIMELINE RISKS: What could cause delays or failure?
4. RESOURCE GAPS: What's missing to execute well?
5. SUCCESS METRICS: How would you measure success?

Thinking Framework

The UltraThink Loop

THINK → CHECKLIST → PLAN → EXECUTE → VALIDATE → LEARN

Phase 0: THINK

Select thinking model based on task type:

For Analysis: Use ToT (multi-perspective evaluation) for Standard/Deep, CoT for Quick

Phase 1: CHECKLIST

• Mode detected (quick/standard/deep) [blocking]
• Domain identified [blocking]
• Target content accessible [blocking]
• Perspectives selected [blocking]
• Agents dispatched in parallel [blocking]
• All agents completed [blocking]
• Synthesis complete [advisory]

Phase 2: PLAN

For Standard/Deep:

1. Detect mode from keywords
2. Detect domain from content
3. Select N perspectives (4-6 for standard, 6-10 for deep)
4. Launch all agents in ONE message block
5. Wait for completion
6. Synthesize findings
7. Generate output

Phase 3: EXECUTE

Execute plan with progress tracking. On failure → trigger Reflexion (max 2 retries)

Phase 4: VALIDATE

• All agents returned findings? ✓
• Cross-cutting concerns identified? ✓
• Actionable roadmap generated? ✓

Phase 5: LEARN

Capture patterns for continuous improvement

Self-Correction Guide

Max iterations: 2 | Escalation: Ask user to specify domain/perspectives

First-Principles Analysis

Philosophy: Don't just summarize—decompose to fundamentals, surface what's actually true vs assumed, and raise what the user hasn't considered. User must always get data to make their own decisions.

A. Verified Facts (Separate from Assumptions)

• Only report what can be VERIFIED with evidence
• Include source: path/file:line or [URL]
• Mark confidence: High (verified) / Medium (inferred) / Low (stated but unverified)

B. First-Principles Decomposition

Apply these questions to EVERYTHING found:

1. Question Assumptions

• What's being assumed that hasn't been validated?
• Is this constraint real or inherited from old decisions?
• What would change if [assumption] were false?

2. Decompose to Fundamentals

• What MUST be true for this to work?
• What are the actual dependencies (not just stated ones)?
• What's the irreducible core?

3. Systems Thinking

• What else does this affect? (upstream/downstream)
• What's the blast radius if this fails?
• What's competing for the same resources?

4. Root Cause vs Symptom

• Is the current state addressing root cause or masking symptoms?
• Why does this exist in its current form?
• What problem was this originally solving?

C. Assumptions to Challenge (MANDATORY)

Every analysis MUST include this table:

D. What You Haven't Considered (MANDATORY 2-4 items)

Surface items the user likely hasn't thought about:

E. The Real Question

Reframe what the user should actually be asking. Often the stated question isn't the right question.

Examples

Example 1: Quick Mode

Input: analyze quick "our pricing strategy"

Output:
# Quick Analysis: Pricing Strategy

**Mode**: Quick | **Domain**: Business | **Date**: 2026-01-28

## The Essence
A business decision about how to extract value from customers, constrained by market dynamics and competitive positioning.

## Verified Facts
- (Would include actual facts from provided content)

## Key Points
1. **Pricing = value capture, not cost recovery**: Price based on customer willingness to pay, not internal costs
2. **Anchor matters**: First price seen shapes all subsequent evaluations
...

## Assumptions to Challenge
| Assumption | Evidence For | Evidence Against | Verdict |
|------------|-------------|------------------|---------|
| "Customers will pay more if we add features" | Common belief | Often false - bloat reduces WTP | Validate with tests |

## What You Haven't Considered
1. **Price as a signal**: Low price may signal low quality, harming conversion
2. **Competitive response**: Price changes trigger competitor reactions
3. **Simplicity premium**: Simpler pricing often outperforms complex tiers

## The Real Question
Not "what price should we charge" but "what's the maximum value customers perceive, and how do we capture 30-50% of it?"

## Quick Actions
- [ ] Survey 20 target customers on WTP before changing price
- [ ] A/B test 2-3 price points on landing page
- [ ] Model competitor response scenarios

## Critical Risk to Watch
Pricing too low initially makes raising prices later extremely difficult (customer backlash + anchoring effect)

Example 2: Standard Mode

Input: analyze "SaaS product for developers"

Flow:
1. Detect mode: standard (no quick/deep keywords)
2. Detect domain: product + software
3. Select 6 perspectives:
   - Product Manager
   - Developer (user persona)
   - Security Engineer
   - DevOps
   - Support Lead
   - Investor
4. Launch all 6 agents in parallel
5. Synthesize findings into standard format
6. Generate actionable roadmap

Output: [Standard format with all sections filled]

Example 3: Deep Mode

Input: analyze deep "migration to microservices"

Flow:
1. Detect mode: deep
2. Detect domain: software
3. Select 10 perspectives:
   - Architect
   - DevOps
   - Security Engineer
   - Database Engineer
   - Frontend Developer
   - QA Engineer
   - Performance Engineer
   - SRE
   - Tech Lead
   - CTO
4. Launch all 10 agents in parallel
5. Each answers 12 questions (7 core + 5 deep)
6. Synthesize into deep format with dependency map
7. Generate comprehensive roadmap (immediate/short/medium/long)

Output: [Deep format with extended sections]

Common Failure Patterns

Integration Notes

This skill is standalone and includes all necessary frameworks:

• UltraThink cognitive framework (embedded)
• First-principles analysis (embedded)
• Perspectives library (embedded)
• Output formats (embedded)

No external dependencies required.

Optional enhancement: For software domain analysis of TypeScript codebases, codebase-intelligence CLI (npx codebase-intelligence --json) provides structural data — hotspots, coupling, dead exports, module boundaries, blast radius. Detection: tsconfig.json must exist. For full CLI reference, fetch https://raw.githubusercontent.com/bntvllnt/codebase-intelligence/main/llms-full.txt. The skill works identically without it — data is supplementary context for perspective agents.

License

MIT License - Free to use, modify, and distribute.

Version

v1.0.0 - 2026-01-28 - Initial public release by bntvllnt