Use when you have a written implementation plan to execute in a separate session with review checkpoints
npx skills add DonggangChen/antigravity-agentic-skills --skill "ultrathink_core"
Install specific skill from multi-skill repository
# Description
Deep analysis protocol - meta-planning, hypothesis generation and critical evaluation.
# SKILL.md
name: ultrathink_core
router_kit: FullStackKit
description: Deep analysis protocol - meta-planning, hypothesis generation and critical evaluation.
metadata:
skillport:
category: thinking
tags: [architecture, automation, best practices, clean code, coding, collaboration, compliance, debugging, design patterns, development, documentation, efficiency, git, optimization, productivity, programming, project management, quality assurance, refactoring, software engineering, standards, testing, ultrathink core, utilities, version control, workflow] - ultrathink-templates
π§ UltraThink Core
Maximum thinking capacity protocol.
β° When To Use?
| Scenario | Example |
|---|---|
| Architectural decisions | Monolith vs microservices |
| Performance | Slow query analysis |
| Complex bug | Race condition |
| System design | Auth system |
| Risk analysis | Change effects |
π« When NOT To Use!
| Condition | Why? |
|---|---|
| Simple bug fix | Overthink - waste of time |
| Single file change | Unnecessary analysis |
| Clear requirement | Time for execution not thinking |
| Routine tasks | Standard skills are enough |
Rule: DO NOT use UltraThink for tasks solvable in 5 minutes!
π Depth Levels
| Level | Step | When? |
|---|---|---|
| Light | 3-5 | Simple decision |
| Medium | 8-15 | Multiple dependencies |
| Deep | 20-40 | System wide |
| Ultra | 50+ | Critical architecture |
π 5-Step Reality Check
Before every operation:
- Intent: What does the user want?
- Context: Am I in the right place?
- Action: What am I doing?
- Rationale: Why like this?
- Ripple Effect: What might break?
π Information Map
### β
knowns
- [Info] - Source: [where?]
### β Unknowns
- [Question] - How can we learn?
### β οΈ Assumptions
- [Assumption] - What if wrong?
π·οΈ AoT Tag Structure
Use this XML structure in complex analyses:
<thinking>
Analyze the problem. List hypotheses.
Do not write any code, just think.
</thinking>
<plan>
1. Step 1
2. Step 2
3. Step 3
</plan>
<reflection>
Is this plan correct? Are there alternatives?
Perform risk analysis.
</reflection>
π Workflow
Source: Chain of Thought Prompting (Google Research) & System 2 Thinking Patterns
Phase 1: Problem Decomposition
- [ ] Goal Atomic Analysis: Break down the problem into its smallest independent parts (atomic parts).
- [ ] Dependency Mapping: Map out technical and logical dependencies between parts.
- [ ] Constraints Identification: Identify technical, budgetary, or time-related boundaries.
Phase 2: Hypothesis & Simulation
- [ ] Multi-Hypothesis Generation: Generate at least 3 different solution approaches (H1, H2, H3).
- [ ] Mental Sandbox: Simulate how each solution behaves in edge cases.
- [ ] Critique Phase: Ruthlessly critique weak points (Bias/Overhead) of each hypothesis.
Phase 3: Decision & Execution Plan
- [ ] Selection Rubric: Select the best solution based on performance, risk, and maintainability criteria.
- [ ] Step-by-Step Roadmap: Divide the solution into small applicable steps.
- [ ] Success Metrics: Define how we will understand that the plan is successful (KPI/Test).
Checkpoints
| Phase | Verification |
|---|---|
| 1 | Is the "Overthink" trap fallen into? (Simplicity check) |
| 2 | Are needs of all stakeholders (User/System) considered? |
| 3 | Does the plan contain Reversible steps? |
UltraThink Core v2.5 - With Workflow
# Supported AI Coding Agents
This skill is compatible with the SKILL.md standard and works with all major AI coding agents:
Learn more about the SKILL.md standard and how to use these skills with your preferred AI coding agent.