Cerebras Coordinator Skill

Description

Coordinates fast code generation via Z.ai glm-4.6 model with CodeSearch pattern learning. Agents use this skill to offload rapid test generation and boilerplate code while building a searchable database of successful patterns.

Key Features

• 🚀 Fast Code Generation: Uses Cerebras API for rapid code creation
• 📚 Pattern Learning: Tracks successful prompts and contexts in CodeSearch
• 🔄 Feedback Loop: Tests generated code and logs results
• 🎯 Agent Coordination: Provides simple interface for agents to coordinate generation tasks
• 📊 Success Metrics: Analyzes and ranks prompt effectiveness

Usage

Basic Usage (Agent Pattern)

# Generate code with automatic testing
./coordinate-generation.sh \
  --agent-id "backend-developer-123" \
  --file-path "src/api_handler.rs" \
  --prompt "Create a REST API handler with authentication" \
  --test-command "cargo test api_handler"

Advanced Usage with Context

# Generate with context files and custom settings
./coordinate-generation.sh \
  --agent-id "frontend-dev-456" \
  --file-path "components/UserProfile.tsx" \
  --prompt "Create React component with TypeScript" \
  --context-files "src/types.ts,src/hooks/useAuth.ts" \
  --test-command "npm test -- --testPathPattern=UserProfile" \
  --model "qwen2.5-coder-32b" \
  --max-attempts 3

Query Successful Patterns

# Find what worked for similar files
./query-patterns.sh \
  --file-type "rs" \
  --pattern "REST API"
  --limit 5

# Get agent-specific successful patterns
./query-patterns.sh \
  --agent-id "backend-developer" \
  --success-rate-threshold 0.8

Architecture

Agent (Coordinator)        Cerebras Coordinator Skill         CodeSearch
        |                           |                              |
        |--- Request Generation --->|                              |
        |                           |--- Store Prompt ------------->|
        |                           |                              |
        |<--- Return Result --------|                              |
        |                           |                              |
        |--- Test Validation ------->|                              |
        |                           |                              |
        |--- Feedback -------------->|--- Log Success/Failure ----->|
        |                           |                              |

When to Use

• ✅ Bulk test generation - generating many test files quickly
• ✅ Boilerplate with patterns - learning from previous successful generations
• ✅ Agent code offloading - when agents need fast, simple code generation
• ✅ Repetitive tasks - migrations, similar components, data models
• ❌ NOT for complex logic, security code, or architectural decisions

Configuration

# Required
export ZAI_API_KEY="your-zai-api-key"  # or CEREBRAS_API_KEY for legacy
export CODESEARCH_INDEX_PATH="./.claude/skills/cfn-codesearch/data"

# Optional
export ZAI_MODEL="glm-4.6"  # Fast, cost-effective model (default)
export COORDINATION_DB_PATH="./.claude/skills/cfn-cerebras-coordinator/generations.db"
export DEFAULT_TEST_TIMEOUT="60"
export MAX_GENERATION_ATTEMPTS="3"

Workflow

1. Agent Request: Agent calls coordinator with generation task
2. Pattern Lookup: Coordinator queries CodeSearch for similar successful patterns
3. Prompt Enhancement: Enhances prompt with successful pattern examples
4. Generation: Sends to Cerebras for code generation
5. Testing: Automatically runs tests on generated code
6. Validation: Checks if tests pass and code compiles
7. Logging: Stores results in CodeSearch for future learning
8. Feedback: Returns result to agent with success metrics

Success Metrics

The system tracks:

• Prompt effectiveness by file type
• Agent-specific success rates
• Context file correlations
• Model performance comparisons
• Test pass/fail rates
• Compilation success rates

This creates a self-improving system that gets better at generating code over time.