MCP Workflow · Prompt Optimization (P1)

Server / tool: user-prompt-optimizer · promptenhancer (params: originalPrompt / optimizedPrompt, optional strings, one per round).

Order vs P0 (Hard Constraint)

• P0 (knowledge retrieval) before P1. If both hit: query_text first, then promptenhancer.
• When user mentions knowledge base / LightRAG / daniel_lightrag, never use P1 instead of or before P0.

Trigger

Requirement is vague, too short, or missing key constraints.

Do Not Trigger

• "Just do it" / "don't optimize"
• Clear single-step (rename field, one command)
• Bug with stack trace / file / repro
• User provided full spec / YAML / API doc
• Read-only tasks (rule review, concept explanation, code review) — aligned with mcp-core negative list

vs brainstorming

Pick one, not both: vague requirement → promptenhancer; multi-option comparison → brainstorming (see superpowers-triggers).

Two-Round Flow

1. Round 1: originalPrompt="<user raw requirement>" → optimization guide (do not end task with guide alone).
2. Round 2: optimizedPrompt="<prompt revised per guide>" → executable final spec.

Fast Path (Default)

Complete both rounds in one turn then implement directly; pause for confirmation only when product tradeoffs, security, or multi-option decisions need user input.

Skip Round 1

When user already provides an optimized prompt, go straight to round 2.