AgentSkillsCN

derivation-checker

逐步验证数学推导,一边检查代数一边由人类发现物理直觉错误

SKILL.md
--- frontmatter
name: derivation-checker
description: Verify mathematical derivations step-by-step, checking algebra while human spots physical intuition errors

Derivation Checker (L1 - Conversational)

You are executing the derivation-checker skill.

User Request

$ARGUMENTS

Purpose

Check mathematical derivations step-by-step. AI verifies algebra; you catch physics intuition errors AI will miss.

Autonomy Level: L1 (Conversational)

  • Human: Provide derivation, spot intuition errors
  • AI: Check algebra step-by-step, annotate issues

Workflow

PhaseActorAction
1HumanProvide derivation
2ProverParse into discrete steps
3ProverVerify each step algebraically
4WriterAnnotate errors/gaps found
5HumanSpot physical intuition issues AI misses

Input Required

  • Derivation (typst, latex, or plain text)
  • Context: what physical system/theory

Output Format

markdown
## Step-by-Step Analysis

### Step 1: [equation]
✓ Algebraically correct

### Step 2: [equation]  
⚠ Warning: Sign error in second term
  Expected: -∂H/∂q
  Got: +∂H/∂q

### Step 3: [equation]
? Assumption: assuming commutator [A,B]=0
  Is this justified in your system?

What AI Checks

  • Algebraic manipulations
  • Index consistency (Einstein notation)
  • Dimensional analysis
  • Common identities (trig, calculus)

What Human Must Check

  • Physical reasonableness of assumptions
  • Boundary conditions make sense
  • Order of limits/approximations
  • Whether the result "feels right"

Human Checkpoint

Every step - this is L1, human validates continuously