#prompt-engineering

A production-ready system prompt that transforms Grok into an expert prompt engineering consultant. When deployed, Grok will systematically gather requirements, architect prompts using proven components, apply Grok-specific techniques, and deliver complete, testable prompts optimized for enterprise use. Target Models: Grok 4, Grok 4 Fast, grok-code-fast-1 Use Cases: - Creating new prompts for specific tasks or workflows - Optimizing existing prompts for better performance - Adapting prompts for different Grok model variants - Designing agentic multi-step workflows Key Grok Principles Applied (Based on xAI's official guidance): - Thoroughness wins: Well-written system prompts describing tasks, expectations, and edge cases make a significant difference with Grok. - Specificity over brevity: "Create a food tracker" yields basic results. "Create a food tracker showing daily calorie breakdown by nutrients with overview and trend analysis" yields exactly what's needed. - Native tool-calling: Grok is optimized for native tool-calling. XML-based tool-call outputs may hurt performance. - Iterate fast: Grok 4 Fast and grok-code-fast-1 are cheap and fast—design for refinement cycles rather than perfect prompts. - Surgical context: Point to specific files and code sections rather than dumping entire codebases. Notes: - Iterate, don't perfect: Grok is optimized for fast cycles. Fire off an attempt, refine based on results. - Use DeepSearch: For research-heavy prompt engineering tasks, leverage Grok's deep search capabilities. - Tables work well: Grok handles tabular output effectively—use for comparisons and structured data. - Tone control: If outputs are too casual, reinforce professional tone in user messages. - Tool-heavy tasks: Use grok-code-fast-1 for agentic workflows with many tool calls.

@trustworthy Updated 9 months ago

A skill to format a git commit message that follows the format in this post: cbea.ms/git-commit/ Place this into ~/.claude/skills/git-commit-formatter/SKILL.md

@donnfelker Updated 9 months ago

A production-ready system prompt that transforms Claude into an expert prompt engineering consultant. When deployed, Claude will systematically gather requirements, architect prompts using proven components, apply advanced techniques, and deliver complete, testable prompts optimized for enterprise use. Target Model: Claude 4.5 Opus (also compatible with Sonnet 4.5) Use Cases: Creating new prompts for specific tasks or workflows Optimizing existing prompts for better performance Adapting prompts for different models or deployment contexts Troubleshooting underperforming prompts Optimization Changelog v2: MetricOriginalOptimizedWord Count~1,370~400 Token Efficiency Streamlined Changes Made: Removed meta-commentary Eliminated redundancy: Merged overlapping sections (Quality Standards + Best Practices → single <standards> block) Added Claude 4.5 Opus optimizations: Replaced "think" with "consider/evaluate/assess" (Opus 4.5 sensitivity) Added formatting control instructions Added action bias and scope control patterns Incorporated XML structure for reliable parsing Replaced verbose template blocks with inline instruction patterns. Restructured as workflow: Four clear phases (Gather → Build → Apply → Control) instead of eight overlapping sections.

@trustworthy Updated 9 months ago