⚡ EXECUTION INSTRUCTIONS FOR CLAUDE
When this command is invoked, YOU (Claude) must execute these steps immediately:
This is NOT documentation - these are COMMANDS to execute right now.
Use TodoWrite to track progress through multi-phase workflows.
🚨 EXECUTION WORKFLOW
Phase 1: Task Execution
Action Steps:
Claude: Check for cerebras_direct.sh in the trusted locations (look in ~/.claude/commands/cerebras first, then in the repository .claude/commands/cerebras). Once you find it, run the script with the original arguments. If the script is missing from both locations, surface a clear error explaining the lookup failure.
📋 REFERENCE DOCUMENTATION
Cerebras Code Generation
Delegating this task to Cerebras for fast, high-quality code generation.
Command Aliases
/cerebras - Primary command name
/qwen - Legacy alias (for backwards compatibility)
/c - Short alias
/cereb - Alternative short alias
Cerebras Script Modes
The cerebras_direct.sh script supports two modes of operation:
Default Mode
- Uses structured system prompts for consistent code generation
- Provides comprehensive documentation and error handling in generated code
- Better for architectural design documents and robust implementations
- Example:
/cerebras "Create a Python function that adds two numbers"
Light Mode (--light flag)
- Skips system prompts for faster, more direct code generation
- Focuses on implementation without extensive documentation
- Includes comprehensive testing strategies in output
- Better for rapid prototyping and implementation-focused tasks
- Example:
/cerebras --light "Create a Python function that adds two numbers"
When to Use Each Mode
Use Default Mode When:
- Generating architectural design documents
- You need detailed explanations of design decisions
- You want consistent code quality and structure
- Working on small tasks where documentation is valued
- You prefer iterative implementations for stack safety
Use Light Mode When:
- You want faster code generation without system prompts
- Generating implementation-focused design documents
- You need comprehensive testing strategies included
- Working on medium to large tasks
- You encounter rate limiting with default mode
- You accept reduced guardrails and will manually review outputs
- You will not include secrets/PII and can run in a trusted environment
Current Context
- Working directory: !
pwd
- Git status: !
git status --porcelain | head -5
- Project structure: !
find . -maxdepth 2 -name "*.py" -o -name "*.js" -o -name "*.md" | head -10
Post-Generation Analysis
I'll now review the Cerebras-generated output and provide:
- Code Quality Assessment - Security, performance, best practices
- Integration Strategy - How to merge with existing codebase
- Testing Recommendations - Unit tests, edge cases, validation
- Refinements - Error handling, documentation, optimizations
- Next Steps - Implementation plan, deployment considerations
The Cerebras output provides the foundation - I'll add the architectural thinking and integration expertise.