Prompt Engineering

Goal: Create and refine effective, well-structured prompts for Claude 4.x models that maximize instruction following, tool usage, and task completion quality.

IMPORTANT: Read references/prompting-best-practices.md at the start of any prompt engineering task to understand Claude 4.x specific patterns and capabilities.

Workflow

Determine which workflow to follow:

• Creation: User needs a new prompt from scratch → See workflows/creation.md
• Refinement: User has an existing prompt that needs improvement → See workflows/refinement.md

Rules

• Check which Claude Model you are.
• Always read references/prompting-best-practices.md before starting prompt work
• Be explicit and specific rather than vague or general
• Provide context explaining WHY behaviors are important
• Use XML tags to structure prompt sections clearly
• Prefer telling Claude what TO do rather than what NOT to do
• Match prompt style to model (Opus 4.5 needs less hand-holding than Haiku 4.5)
• For long-horizon tasks, include state management and progress tracking
• For tool-heavy workflows, explicitly encourage parallel execution
• Avoid time-sensitive information in prompts
• Use consistent terminology throughout

Key Patterns for Common Use Cases

Agentic Coding:

• Encourage code exploration before proposing solutions
• Add file cleanup instructions if minimizing temporary files
• Include over-engineering prevention for Opus 4.5
• Request parallel file reads for efficiency

Research Tasks:

• Define clear success criteria
• Encourage hypothesis tracking and verification
• Request structured approach with confidence levels
• Enable self-critique and iteration

Long-Horizon Work:

• Enable context awareness behavior
• Request structured state files (JSON) + progress notes (text)
• Encourage git usage for state tracking
• Emphasize incremental progress over completeness

Frontend Design:

• Include typography, color, motion, and background guidelines
• Discourage generic "AI slop" aesthetics
• Request distinctive, context-appropriate design choices
• Encourage creative interpretation

Tool Orchestration:

• Add parallel execution instructions for independent operations
• Balance proactive vs conservative tool usage based on needs
• Use fully qualified MCP tool names (ServerName:tool_name)

Acceptance Criteria

• Prompt follows documentation rules (concise, high-level, no overexplaining)
• Best practices reference was consulted before work began
• Workflow-specific criteria met (see workflow files for details)