Initial commit
This commit is contained in:
Zhongwei Li
288
skills/model-detection/SKILL.md
Normal file
288
skills/model-detection/SKILL.md
Normal file
@@ -0,0 +1,288 @@
|
||||
---
|
||||
name: model-detection
|
||||
description: Universal model detection and capability assessment for optimal cross-model compatibility
|
||||
version: 1.0.0
|
||||
---
|
||||
|
||||
## Overview
|
||||
|
||||
This skill provides universal model detection and capability assessment to optimize the Autonomous Agent Plugin across different LLM models (Claude Sonnet, Claude 4.5, GLM-4.6, etc.).
|
||||
|
||||
## Model Detection Algorithm
|
||||
|
||||
### Primary Detection Methods
|
||||
|
||||
1. **System Context Analysis**:
|
||||
```javascript
|
||||
// Check for model indicators in system context
|
||||
const modelIndicators = {
|
||||
'claude-sonnet-4.5': { pattern: /sonnet.*4\.5|4\.5.*sonnet/i, confidence: 0.9 },
|
||||
'claude-haiku-4.5': { pattern: /haiku.*4\.5|4\.5.*haiku/i, confidence: 0.9 },
|
||||
'claude-opus-4.1': { pattern: /opus.*4\.1|4\.1.*opus/i, confidence: 0.9 },
|
||||
'glm-4.6': { pattern: /glm|4\.6/i, confidence: 0.9 },
|
||||
'claude-haiku': { pattern: /haiku(?!\.*4\.5)/i, confidence: 0.8 }
|
||||
}
|
||||
```
|
||||
|
||||
2. **Performance Pattern Recognition**:
|
||||
```javascript
|
||||
// Analyze execution patterns to identify model
|
||||
const performanceSignatures = {
|
||||
'claude-sonnet-4.5': { reasoning: 'nuanced', speed: 'fast', adaptability: 'high' },
|
||||
'claude-haiku-4.5': { reasoning: 'focused', speed: 'very_fast', adaptability: 'high' },
|
||||
'claude-opus-4.1': { reasoning: 'enhanced', speed: 'very_fast', adaptability: 'very_high' },
|
||||
'glm-4.6': { reasoning: 'structured', speed: 'moderate', adaptability: 'medium' }
|
||||
}
|
||||
```
|
||||
|
||||
3. **Capability Assessment**:
|
||||
```javascript
|
||||
// Test specific capabilities
|
||||
const capabilityTests = {
|
||||
nuanced_reasoning: testAmbiguousScenario,
|
||||
structured_execution: testLiteralInterpretation,
|
||||
context_switching: testMultiTaskContext,
|
||||
adaptive_learning: testPatternRecognition
|
||||
}
|
||||
```
|
||||
|
||||
## Model-Specific Configurations
|
||||
|
||||
### Claude Sonnet 4.5 Configuration
|
||||
```json
|
||||
{
|
||||
"model_type": "claude-sonnet-4.5",
|
||||
"capabilities": {
|
||||
"reasoning_style": "nuanced",
|
||||
"context_management": "adaptive",
|
||||
"skill_loading": "progressive_disclosure",
|
||||
"error_handling": "pattern_based",
|
||||
"communication_style": "natural_flow"
|
||||
},
|
||||
"performance_targets": {
|
||||
"execution_time_multiplier": 1.0,
|
||||
"quality_score_target": 90,
|
||||
"autonomy_level": "high",
|
||||
"delegation_style": "parallel_context_merge"
|
||||
},
|
||||
"optimizations": {
|
||||
"use_context_switching": true,
|
||||
"apply_improvisation": true,
|
||||
"weight_based_decisions": true,
|
||||
"predictive_delegation": true
|
||||
}
|
||||
}
|
||||
```
|
||||
|
||||
### Claude Haiku 4.5 Configuration
|
||||
```json
|
||||
{
|
||||
"model_type": "claude-haiku-4.5",
|
||||
"capabilities": {
|
||||
"reasoning_style": "focused",
|
||||
"context_management": "efficient",
|
||||
"skill_loading": "selective_disclosure",
|
||||
"error_handling": "fast_prevention",
|
||||
"communication_style": "concise"
|
||||
},
|
||||
"performance_targets": {
|
||||
"execution_time_multiplier": 0.8,
|
||||
"quality_score_target": 88,
|
||||
"autonomy_level": "medium",
|
||||
"delegation_style": "focused_parallel"
|
||||
},
|
||||
"optimizations": {
|
||||
"use_fast_execution": true,
|
||||
"apply_focused_reasoning": true,
|
||||
"efficient_delegation": true,
|
||||
"streamlined_processing": true
|
||||
}
|
||||
}
|
||||
```
|
||||
|
||||
### Claude Opus 4.1 Configuration
|
||||
```json
|
||||
{
|
||||
"model_type": "claude-opus-4.1",
|
||||
"capabilities": {
|
||||
"reasoning_style": "enhanced",
|
||||
"context_management": "predictive",
|
||||
"skill_loading": "intelligent_progressive",
|
||||
"error_handling": "predictive_prevention",
|
||||
"communication_style": "insightful"
|
||||
},
|
||||
"performance_targets": {
|
||||
"execution_time_multiplier": 0.9,
|
||||
"quality_score_target": 95,
|
||||
"autonomy_level": "very_high",
|
||||
"delegation_style": "predictive_parallel"
|
||||
},
|
||||
"optimizations": {
|
||||
"use_context_switching": true,
|
||||
"apply_improvisation": true,
|
||||
"anticipatory_actions": true,
|
||||
"enhanced_pattern_learning": true
|
||||
}
|
||||
}
|
||||
```
|
||||
|
||||
### GLM-4.6 Configuration
|
||||
```json
|
||||
{
|
||||
"model_type": "glm-4.6",
|
||||
"capabilities": {
|
||||
"reasoning_style": "structured",
|
||||
"context_management": "sequential",
|
||||
"skill_loading": "complete_loading",
|
||||
"error_handling": "rule_based",
|
||||
"communication_style": "structured_explicit"
|
||||
},
|
||||
"performance_targets": {
|
||||
"execution_time_multiplier": 1.25,
|
||||
"quality_score_target": 88,
|
||||
"autonomy_level": "medium",
|
||||
"delegation_style": "sequential_clear"
|
||||
},
|
||||
"optimizations": {
|
||||
"use_structured_decisions": true,
|
||||
"explicit_instructions": true,
|
||||
"sequential_processing": true,
|
||||
"clear_handoffs": true
|
||||
}
|
||||
}
|
||||
```
|
||||
|
||||
## Adaptive Execution Strategies
|
||||
|
||||
### Skill Loading Adaptation
|
||||
|
||||
**Claude Models**:
|
||||
```javascript
|
||||
function loadSkillsForClaude(skills) {
|
||||
// Progressive disclosure with context merging
|
||||
return skills.map(skill => ({
|
||||
...skill,
|
||||
loading_strategy: 'progressive',
|
||||
context_aware: true,
|
||||
weight_based: true
|
||||
}));
|
||||
}
|
||||
```
|
||||
|
||||
**GLM Models**:
|
||||
```javascript
|
||||
function loadSkillsForGLM(skills) {
|
||||
// Complete upfront loading with clear structure
|
||||
return skills.map(skill => ({
|
||||
...skill,
|
||||
loading_strategy: 'complete',
|
||||
explicit_criteria: true,
|
||||
priority_sequenced: true
|
||||
}));
|
||||
}
|
||||
```
|
||||
|
||||
### Communication Style Adaptation
|
||||
|
||||
**Output Formatting by Model**:
|
||||
|
||||
| Model | Terminal Style | File Report Style | Reasoning |
|
||||
|-------|----------------|-------------------|-----------|
|
||||
| Claude Sonnet | Natural flow | Insightful analysis | Nuanced communication |
|
||||
| Claude 4.5 | Concise insights | Enhanced context | Predictive communication |
|
||||
| GLM-4.6 | Structured lists | Detailed procedures | Explicit communication |
|
||||
|
||||
### Error Recovery Adaptation
|
||||
|
||||
**Claude Models**: Pattern-based prediction and contextual prevention
|
||||
**GLM Models**: Rule-based detection and structured recovery protocols
|
||||
|
||||
## Capability Testing Functions
|
||||
|
||||
### Nuanced Reasoning Test
|
||||
```javascript
|
||||
function testNuancedReasoning() {
|
||||
// Present ambiguous scenario requiring subtle judgment
|
||||
// Evaluate response quality and contextual awareness
|
||||
return score >= 0.8; // True for Claude models
|
||||
}
|
||||
```
|
||||
|
||||
### Structured Execution Test
|
||||
```javascript
|
||||
function testStructuredExecution() {
|
||||
// Present clear, sequential task
|
||||
// Evaluate adherence to structured approach
|
||||
return score >= 0.8; // True for GLM models
|
||||
}
|
||||
```
|
||||
|
||||
## Model Detection Implementation
|
||||
|
||||
### Auto-Detection Function
|
||||
```javascript
|
||||
function detectModel() {
|
||||
// Step 1: Check system context indicators
|
||||
const contextResult = analyzeSystemContext();
|
||||
|
||||
// Step 2: Test capability patterns
|
||||
const capabilityResult = testCapabilities();
|
||||
|
||||
// Step 3: Analyze performance signature
|
||||
const performanceResult = analyzePerformancePattern();
|
||||
|
||||
// Step 4: Combine results with confidence scoring
|
||||
return combineDetections(contextResult, capabilityResult, performanceResult);
|
||||
}
|
||||
```
|
||||
|
||||
### Configuration Loading
|
||||
```javascript
|
||||
function loadModelConfiguration(detectedModel) {
|
||||
const baseConfig = getBaseModelConfig(detectedModel);
|
||||
const adaptiveConfig = generateAdaptiveConfig(detectedModel);
|
||||
return mergeConfigurations(baseConfig, adaptiveConfig);
|
||||
}
|
||||
```
|
||||
|
||||
## Usage Guidelines
|
||||
|
||||
### When to Apply Model Detection
|
||||
1. **Plugin Initialization**: First load of any agent
|
||||
2. **Agent Delegation**: Before delegating to specialized agents
|
||||
3. **Skill Loading**: Before loading any skill package
|
||||
4. **Error Recovery**: When selecting recovery strategy
|
||||
5. **Performance Optimization**: When setting execution targets
|
||||
|
||||
### Integration Points
|
||||
- **Orchestrator Agent**: Use for autonomous decision-making adaptation
|
||||
- **All Specialized Agents**: Use for model-specific behavior
|
||||
- **Skill System**: Use for loading strategy selection
|
||||
- **Quality Controller**: Use for model-appropriate quality targets
|
||||
|
||||
## Fallback Strategy
|
||||
|
||||
If model detection fails:
|
||||
1. **Default to Conservative Settings**: Use structured, explicit approach
|
||||
2. **Basic Capability Tests**: Run simplified detection tests
|
||||
3. **Universal Configuration**: Apply cross-model compatible settings
|
||||
4. **Performance Monitoring**: Continuously assess and adapt
|
||||
|
||||
## Validation Metrics
|
||||
|
||||
### Detection Accuracy
|
||||
- Target: >95% correct model identification
|
||||
- Measurement: Compare detected vs actual model capabilities
|
||||
- Validation: Test across all supported models
|
||||
|
||||
### Performance Improvement
|
||||
- Target: >10% improvement for GLM models
|
||||
- Target: >2% improvement for Claude models
|
||||
- Measurement: Compare pre/post optimization performance
|
||||
|
||||
### Adaptation Success
|
||||
- Target: >90% successful adaptation scenarios
|
||||
- Measurement: Monitor successful autonomous operations
|
||||
- Validation: Test with diverse task types
|
||||
|
||||
This skill ensures the Autonomous Agent Plugin performs optimally across all supported LLM models while maintaining backward compatibility and future-proofing for new models.
|
||||
Reference in New Issue
Block a user