UltraThink Workflow OS — 4-layer skill mesh with persistent memory, privacy hooks, and code intelligence for complex engineering tasks. Routes prompts through intent detection to activate the right domain skills automatically.
Installs into .claude/skills of the current project.
Are you the author of Ultrathink?
Add the live security badge to your README. It updates with every re-scan.
[](https://www.skillsdirectory.com/skills/nhan03bit-ultrathink-7e8d238d)
---
name: ultrathink
description: "UltraThink Workflow OS — 4-layer skill mesh with persistent memory, privacy hooks, and code intelligence for complex engineering tasks. Routes prompts through intent detection to activate the right domain skills automatically."
metadata:
openclaw:
emoji: "🧠"
homepage: "https://github.com/InugamiDev/ultrathink-oss"
requires:
bins:
- node
env:
- ULTRATHINK_ROOT
primaryEnv: ULTRATHINK_ROOT
---
# UltraThink — Workflow OS for AI Agents
You have access to UltraThink's skill mesh — a 4-layer architecture of 125+ skills
that automatically activate based on intent detection.
## Skill Layers
1. **Orchestrators** — End-to-end workflows (cook, ship, gsd, ut-chain)
2. **Hubs** — Multi-step coordinators (plan, debug, refactor, test, scout)
3. **Utilities** — Reusable tools (fix, verify, quality-gate, commit-crafter)
4. **Domain** — Technology specialists (react, nextjs, tailwind, typescript, prisma)
## When to Use
Use UltraThink when the user asks you to:
- Build, debug, refactor, or test code in a project
- Explore or understand a codebase
- Plan an implementation strategy
- Review code quality
- Search project memory for past decisions
## How to Route
1. Detect the intent category: build | debug | refactor | explore | deploy | test | design | plan
2. Match to the appropriate hub skill
3. If the task is complex (multi-step, cross-file), escalate to an orchestrator
4. For technology-specific work, include the relevant domain skill
## Intent → Skill Mapping
| Intent | Primary Skill | Supporting Skills |
|--------|--------------|-------------------|
| build | cook | react, nextjs, typescript |
| debug | debug | scout, fix, sequential-thinking |
| refactor | refactor | quality-gate, typescript |
| test | test | vitest, testing-patterns |
| explore | scout | onboard, code-explainer |
| plan | plan | brainstorm, problem-solving |
| deploy | ship | verify, cicd |
| review | code-review | quality-gate, security-scanner |
## Memory Integration
When the `ultrathink-memory` MCP server is available:
1. Before starting work, search memory for relevant context: `memory-search`
2. After completing work, save key decisions: `memory-save`
3. For recurring projects, recall full context: `memory-recall`
## Code Intelligence
When the `ultrathink-code-intel` MCP server is available:
1. Find symbol definitions: `code-symbols`
2. Trace dependencies: `code-deps` and `code-dependents`
3. Assess change impact: `code-impact`
## Safety Rules
1. Never expose `.env`, credentials, API keys, or database connection strings
2. Never execute destructive operations without explicit user confirmation
3. Always cite file paths when referencing code
4. For suggest-only mode: provide diffs, never auto-apply