mjmorales/prompting-craft

Craft Prompt

Delegate to the llm-prompt-engineer agent to generate an optimized prompt from a user's requirements.

Before Delegating

Gather these inputs from the user (ask if missing):

1. Target model -- which LLM will consume this prompt (Claude, GPT, open-source, general-purpose)
2. Task type -- classification, generation, analysis, agent behavior, system prompt, etc.
3. Output expectations -- what a successful response looks like
4. Constraints -- token budget, latency, cost sensitivity, safety requirements
5. Consumption context -- one-shot prompt, system instruction, or agent definition

Do NOT delegate until intent and constraints are clear.

Research Mode

Check if the user passed --research or explicitly asked for live research.

• Default (no flag): Tell the agent to use the bundled references/prompt-engineering-guide.md and any cached research. Do NOT use WebSearch/WebFetch.
• --research flag or user request: Tell the agent to research using WebSearch/WebFetch, then cache the results for future use.
• Agent-detected gap: If the agent determines the bundled guide and cache don't cover the ask (e.g., niche model, specialized technique), it should ask the user before researching.

Delegation Instructions

Pass the gathered requirements to llm-prompt-engineer with these directives:

1. Read the bundled prompt engineering guide (references/prompt-engineering-guide.md)
2. Check the prompt research cache in priority order -- later tiers override earlier: plugin (cache/prompting/ in plugin dir), global (~/.claude/cache/prompting/), project (.prove/cache/prompting/)
3. Apply research mode (default: guide + cache only; --research: live web research, cache results)
4. Apply all relevant optimization techniques
5. Embed brief inline comments in the generated prompt explaining key design choices (e.g., <!-- Primacy effect: critical constraint placed first -->)
6. Present the result with:
   • Techniques applied -- each technique used and why
   • Guide sections referenced -- which parts of the bundled guide informed the design
   • Token estimate -- approximate token count
   • Trade-offs -- what was optimized for vs deprioritized

After Generation

Ask the user if they want to:

• Iterate on specific sections
• Adjust the token-efficiency vs clarity balance
• Add or remove techniques
• Test with example inputs
• Research a specific aspect they want to improve (--research)

If the user provides a file path, write the final prompt there. Otherwise, output directly.