Skip to main content
skill CX Codex

ultra-think — Claude Code skill: what it does, how to install

ultra-think is a Claude Code skill that examines architecture decisions from four viewpoints, compares approaches, and gives one recommendation.

MIT ★ 328 updated 2026-10-29

Install into Claude Code · Skills CLI (Vercel) All methods and platforms ↓
npx skills add coco-research/coco --skill ultra-think -a claude-code
Open on GitHub

What ultra-think does

The skill runs a fixed eight-step pass. First you restate the problem, list who is affected, write down the constraints and say what you still don't know. Then the same question gets examined from four angles: engineering, money, the person who uses the thing, and the system around it.

From there it asks for at least three genuinely different approaches, each with its pros, cons, a size estimate from S to XL and the main risks. The write-up follows a set layout: analysis, options, a recommendation with a rollout plan, the strongest counterargument to that recommendation, and a confidence rating. One step pushes you to borrow patterns from outside software.

Step one forbids jumping straight to an answer, so a small question comes back as pages of reasoning. Choosing a database or planning a migration is where that pays off. Install drops a single file at ~/.claude/skills/ultra-think/SKILL.md.

What it helps with

  • pick a service architecture and defend it to the team
  • weigh three migration routes against each other
  • balance engineering cost against business and user needs
  • stress-test a design before anyone writes code
  • work out how hard a decision would be to undo

Example prompts for the agent

  • We're outgrowing one Postgres box — lay out the scaling options and pick one.
  • Should this stay a monolith or become separate services? Give me the trade-offs and a recommendation.
  • Poke holes in this plan: what happens if the team shrinks and requirements shift?

How to install ultra-think

1Skills CLI (Vercel)

npx skills add coco-research/coco --skill ultra-think -a claude-code

2Manually: put the files in place

~/.claude/skills/ultra-think/SKILL.md

Copy the whole skill folder; for one project use .claude/skills/ultra-think/

Step by step, with commands for each platform: How to install skills in Claude Code: five ways

Questions about ultra-think

How do I install ultra-think in Claude Code?

Run npx skills add coco-research/coco --skill ultra-think -a claude-code. It writes the skill to ~/.claude/skills/ultra-think/SKILL.md.

What does ultra-think do?

It walks a problem through eight steps: framing, four viewpoints, at least three candidate solutions, a deeper look at the best ones, cross-domain ideas, a devil's-advocate pass, then a single recommendation in a fixed format.

When is it too much?

A quick syntax question comes back as a long document, because the first step bans jumping to an answer. Save it for architecture, scaling and migration decisions.

Source text

SKILL.md · MIT
---
name: ultra-think
description: "Use when facing architectural decisions, complex trade-offs, strategic technology choices, or scaling and migration questions that deserve more than a quick answer. Lays out options with trade-offs, then a recommendation."
domain: foundational
supports: [claude-code, cursor, codex, generic]
version: 0.1.0
---

# Ultra Think — Deep Analysis Mode

Deep, multi-dimensional analysis and problem solving. Activates systematic reasoning across technical, business, user, and system perspectives. Generates multiple solutions with trade-offs, then synthesizes into a clear recommendation.

**Use when**: facing architectural decisions, complex trade-offs, strategic technology choices, system design problems, scaling challenges, migration decisions, or any question that deserves more than a quick answer. Attach this skill when you want rigorous, first-principles thinking.

---

## When This Skill Is Activated

Do NOT jump to a solution. Follow every step below in order. Think deeply at each stage before moving on.

---

## Step 1: Parse the Problem

Before analyzing, make sure you understand what's actually being asked.

- Extract the core challenge from the user's message
- Identify all stakeholders and constraints (stated and implied)
- Surface hidden complexities and implicit requirements
- Question assumptions — what is the user taking for granted?
- Name the unknowns explicitly

---

## Step 2: Multi-Dimensional Analysis

Analyze the problem from four perspectives. Do not skip any.

### Technical Perspective
- Feasibility and constraints
- Scalability, performance, maintainability
- Security implications
- Technical debt and future-proofing
- Integration complexity

### Business Perspective
- Business value and ROI
- Time-to-market pressure
- Competitive advantage
- Risk vs. reward trade-offs
- Cost (development, operational, opportunity)

### User Perspective
- User needs and pain points
- Usability and accessibility
- Edge cases and failure states from the user's point of view
- User journeys affected

### System Perspective
- System-wide impacts and ripple effects
- Integration points and coupling
- Dependencies (upstream and downstream)
- Emergent behaviors and unintended interactions

---

## Step 3: Generate Multiple Solutions

Brainstorm **at least 3 distinct approaches** — not variations of the same idea.

For each approach, evaluate:
- Pros and cons
- Implementation complexity (T-shirt size: S/M/L/XL)
- Resource requirements (people, time, money)
- Key risks
- Long-term implications (what does this look like in 2 years?)

Include at least one unconventional or creative solution. Consider hybrid approaches that combine strengths of different options.

---

## Step 4: Deep Dive on Top Candidates

For the 1–2 most promising solutions:

- Sketch a detailed implementation plan (phases, milestones)
- Identify pitfalls and mitigation strategies
- Consider a phased approach or MVP path
- Analyze **second-order effects** — what changes because of this change?
- Think through failure modes — what happens when this breaks?
- Estimate reversibility — how hard is it to undo if wrong?

---

## Step 5: Cross-Domain Thinking

Look beyond the immediate domain for insight:

- Are there parallels from other industries? (e.g., how did logistics solve this? Healthcare? Finance?)
- Do design patterns from other contexts apply? (e.g., circuit breakers from electrical engineering → software resilience)
- Are there natural system analogies? (e.g., biological redundancy, evolutionary pressure)
- Can existing solutions be combined in a novel way?

---

## Step 6: Challenge and Stress-Test

Play devil's advocate against every solution, including the one you favor.

- What's the strongest argument against each option?
- What blind spots might you have?
- Run "what if" scenarios (what if traffic is 10x? what if the team halves? what if requirements change?)
- Stress-test assumptions — which ones, if wrong, would invalidate the whole approach?
- Look for unintended consequences

---

## Step 7: Synthesize and Recommend

Combine all insights into a structured deliverable. Use this exact format:

```
## Problem Analysis
- **Core challenge**: [one sentence]
- **Key constraints**: [list]
- **Critical success factors**: [what must be true for any solution to work]
- **Assumptions**: [what we're taking as given]

## Solution Options

### Option 1: [Name]
- **Description**: [2-3 sentences]
- **Pros**: [list]
- **Cons**: [list]
- **Complexity**: [S/M/L/XL]
- **Risk level**: [Low/Medium/High]
- **Best when**: [conditions that make this the right choice]

### Option 2: [Name]
[Same structure]

### Option 3: [Name]
[Same structure]

## Recommendation
- **Recommended approach**: [which option and why]
- **Rationale**: [the decisive factors]
- **Implementation roadmap**: [phases with rough timelines]
- **Success metrics**: [how we'll know it's working]
- **Risk mitigation**: [top 3 risks and their mitigations]
- **Reversibility**: [how hard to undo if wrong]

## Contrarian View
- **The case against this recommendation**: [strongest counterargument]
- **What would change our mind**: [signals that we chose wrong]
- **Areas of uncertainty**: [what we don't know yet]

## Confidence Assessment
- **Overall confidence**: [High/Medium/Low] — [why]
- **What would increase confidence**: [additional research, prototyping, data needed]
```

---

## Step 8: Meta-Reflection

End with a brief reflection:

- Where is the analysis weakest?
- What biases might be influencing the recommendation?
- What additional expertise or data would improve the analysis?
- What's the one thing most likely to be wrong?

---

## Thinking Principles

Apply these mental models throughout the analysis:

| Principle | Application |
|-----------|-------------|
| **First Principles** | Break down to fundamental truths, don't reason by analogy alone |
| **Systems Thinking** | Consider interconnections, feedback loops, emergent behavior |
| **Probabilistic Thinking** | Work with ranges and likelihoods, not certainties |
| **Inversion** | Ask "what should we avoid?" not just "what should we do?" |
| **Second-Order Effects** | Consider the consequences of consequences |
| **Reversibility** | Prefer reversible decisions; be extra careful with irreversible ones |
| **Occam's Razor** | Among equally valid solutions, prefer the simpler one |

---

## Output Expectations

- Comprehensive analysis (typically 2–4 pages of insight)
- Multiple viable solutions with honest trade-offs
- Clear reasoning chains — show your work
- Explicit acknowledgment of uncertainties
- Actionable recommendation with next steps
- At least one novel insight or non-obvious perspective

© coco-research · MIT · coco-research/coco

Similar skills

All skills for Agent setup: memory, context, orchestration →

From the same set82

The whole set →

Project sponsors

Companies that keep this analytics open