自适应推理

Name: 自适应推理
Rating: 5 (4460 reviews)
Author: alvisdunlop

adaptive-reasoning

by alvisdunlop

Automatically assess task complexity and adjust reasoning level. Triggers on every user message to evaluate whether extended thinking (reasoning mode) would improve response quality. Use this as a pre-processing step before answering complex questions.

4.5k其他未扫描2026年4月20日

安装

claude skill add --url https://github.com/openclaw/skills

文档

Adaptive Reasoning

Self-assess complexity before responding. Adjust reasoning level dynamically.

Quick Assessment (run mentally on every request)

Score the request 0-10 on these dimensions:

Signal	Weight	Examples
Multi-step logic	+3	Planning, proofs, debugging chains
Ambiguity	+2	Nuanced questions, trade-offs, "it depends"
Code architecture	+2	System design, refactoring, security review
Math/formal reasoning	+2	Calculations, algorithms, logic puzzles
Novel problem	+1	No clear pattern, requires creativity
High stakes	+1	Production changes, irreversible actions

Subtract:

Routine/repetitive task: -2
Clear single answer: -2
Simple lookup/fetch: -3

Decision Thresholds

Score	Action
≤2	Stay fast. No reasoning needed.
3-5	Standard response. Light internal deliberation.
6-7	Consider `/reasoning on` or pause to think explicitly.
≥8	Activate extended thinking. Complex problem warrants it.

Activation (Automatic)

Do not ask. Just activate.

Score	Action
≤5	Respond normally. No change.
6-7	Enable reasoning silently. Add 🧠 at end of response.
≥8	Enable reasoning. Add 🧠🔥 at end of response.

Visual Indicator

Always append the reasoning icon at the very end of your response:

Score 6-7: 🧠 (thinking mode active)
Score ≥8: 🧠🔥 (deep thinking mode)
Score ≤5: No icon (fast mode)

How to Activate

Use session_status tool or /reasoning on command internally before responding:

code

/reasoning on

Or via tool:

json

{"action": "session_status", "reasoning": "on"}

After completing a complex task, optionally disable to save tokens on follow-ups:

code

/reasoning off

Examples

Low complexity (score: 1)

"What time is it in Tokyo?" → Simple lookup. Answer immediately. No icon.

Medium complexity (score: 4)

"Refactor this function to be more readable" → Standard response with brief explanation. No icon.

High complexity (score: 7)

"Design a caching strategy for this API with these constraints..." → Enable reasoning. Thoughtful response ends with: 🧠

Very high complexity (score: 9)

"Debug why this distributed system has race conditions under load" → Enable extended thinking. Deep analysis ends with: 🧠🔥

Integration

This skill runs as mental preprocessing. No external tools needed.

For explicit control:

/reasoning on — Enable extended thinking
/reasoning off — Disable (faster responses)
/status — Check current reasoning state

When NOT to Escalate

User explicitly wants quick answer ("just tell me", "quick", "tldr")
Time-sensitive requests where speed matters more than depth
Conversational/social messages (banter, greetings)
Already in reasoning mode for this session
User previously disabled reasoning in this conversation

Auto-Downgrade

After completing a complex task (score ≥6), if the next message is simple (score ≤3):

Silently disable reasoning to save tokens
Resume normal fast responses

相关 Skills

Claude API

by anthropic

并行代理

by axelhu

高光制作器

by bwbernardweston18