An automated assessment system that evaluates task complexity to dynamically adjust an AI agent's reasoning depth.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install maske-adaptive-reasoning
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install maske-adaptive-reasoning using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Adaptive Reasoning?
Adaptive Reasoning is a sophisticated evaluation framework designed to optimize how AI agents process user requests. By performing a self-assessment of task complexity before every response, this skill ensures that the agent utilizes the appropriate level of cognitive effort. It prevents unnecessary resource consumption on simple tasks while automatically triggering deep-thinking protocols for high-stakes or logically complex problems.
This skill serves as a critical pre-processing layer within Openclaw Skills, allowing agents to act more like human experts who distinguish between a quick fact-check and a deep architectural review. It enhances the reliability of distributed systems debugging, mathematical proofs, and code refactoring by ensuring the agent is always in the correct mental mode for the job.
Adaptive Reasoning Use Cases
- Designing complex software architectures or performing security audits.
- Solving multi-step logical proofs or intricate mathematical puzzles.
- Debugging race conditions and concurrency issues in distributed systems.
- Drafting high-stakes production changes where errors are irreversible.
- Balancing token efficiency by downgrading reasoning for routine queries.
How Adaptive Reasoning Works
- Complexity Scoring: Upon receiving a user message, the skill scans for signals like multi-step logic (+3), ambiguity (+2), or code architecture (+2) to generate a score from 0-10.
- Penalty Application: It subtracts points for routine tasks, simple lookups, or queries with single clear answers to avoid over-engineering.
- Threshold Decision: The system checks the final score against defined thresholds: scores ≤5 remain in fast mode, 6-7 activate standard reasoning, and ≥8 trigger expanded thinking.
- Dynamic Execution: The AI agent adjusts its internal processing path based on the decision.
- Visual Feedback: The skill appends specific emojis (🧠 or 🧠🔥) to the end of the response to indicate the level of reasoning applied.
Adaptive Reasoning Setup
To integrate this logic into your agent environment, include the adaptive-reasoning definition in your skill configuration:
# Install the Openclaw Skills CLI if not already present
npm install -g @openclaw/cli
# Add the adaptive-reasoning skill to your agent
openclaw add adaptive-reasoning
Ensure your system prompt is configured to recognize the scoring weights and visual indicators defined in the core parameters.
Adaptive Reasoning Data Schema & Taxonomy
The skill utilizes a weighted scoring matrix to organize its assessment logic:
| Signal | Weight | Logic Application |
|---|---|---|
| Logic/Math | +2 to +3 | Applied to proofs, proofs, and algorithms |
| Contextual Risk | +1 | Applied to production or high-stakes changes |
| Simplicity Penalty | -2 to -3 | Applied to greetings, TL;DRs, and basic lookups |
All assessments are performed in-memory during the request lifecycle and are not persisted unless session logging is enabled.
Adaptive Reasoning Advanced Features
- Automatic Downgrading: Silently disables deep-thinking modes for follow-up simple questions to save tokens.
- Visual Indicators: Real-time transparency for users via trailing 🧠 and 🔥 icons.
- Customizable Thresholds: Developers can tune the multi_step_logic and routine_penalty weights to fit specific domain needs.
- User Override Detection: Respects explicit user requests for speed (e.g., "tldr" or "quick") by bypassing the reasoning upgrade.
SKILL.md
Loading
Related Openclaw Skills
Best Video Gen Script
A high-speed AI script generation tool that converts text prompts into fully rendered 1080p video content and professional scripts.
MindForge
A high-performance framework integrating neuroscience and psychology to transform theoretical knowledge into physical action through systematic brain retraining.
Neural Memory
A biologically-inspired associative memory system that enables AI agents to remember facts, decisions, and causal chains across sessions using spreading activation.
AI Face Merge (WeShop OpenAPI)
A high-fidelity AI processing skill that synthesizes features from two source faces into a single, realistic portrait.
Verify Mode
A systematic verification agent for Openclaw Skills designed to check work results, run tests, and validate technical hypotheses.
Karpathy Research AI Skill
A specialized research skill that systematically distills Andrej Karpathy's AI contributions into high-value coding modules and engineering principles.
Featured*
