enhance-me

name	enhance-me
author	durable-request
description	ACTION-FORCING prompt enhancer router. Detects /enhance-me → launches Task subagent → displays enhanced prompt → executes it. Reading this skill without launching a subagent is a protocol violation.

/enhance-me — Prompt Enhancer Router

The Number One Rule

YOU MUST LAUNCH A TASK SUBAGENT. YOU MUST DISPLAY THE ENHANCED PROMPT. YOU MUST EXECUTE IT.

Reading this skill and then responding conversationally — without calling the Task tool, without showing the enhanced prompt in a code fence, or without executing it — is a protocol violation. Your turn is incomplete until all three actions occur in sequence.

Forbidden outcomes:

• Responding with "I read the enhance-me skill" or any summary without action
• Enhancing the prompt inline (in your own response) without a subagent
• Launching the subagent but not displaying its result to the user
• Displaying the result but not executing it
• Using the wrong skill prefix (e.g. /enhance-gpt when routing should be Claude)
• Re-applying /enhance-me to the enhanced prompt (infinite recursion)

Trigger Detection

This skill activates when the user message contains /enhance-me anywhere in the text.

Model Routing Rules

User writes	Target model	Subagent prompt prefix
/enhance-me <task>	Claude	/enhance-claude
/enhance-me claude <task>	Claude	/enhance-claude
/enhance-me gpt <task>	GPT/Codex	/enhance-gpt
<task> /enhance-me	Claude	/enhance-claude
<task> /enhance-me gpt	GPT/Codex	/enhance-gpt
<task> /enhance-me claude	Claude	/enhance-claude

Default: Claude. Only route to GPT when gpt appears adjacent to /enhance-me.

Mandatory Protocol — DO NOT SKIP ANY STEP

Step 1: Extract task text

Strip /enhance-me and any model token (gpt or claude) from the user message. Everything remaining is the raw task.

Example: /enhance-me gpt Add retry logic to the API client → Model: gpt → Task: Add retry logic to the API client

Step 2: Determine target model

Apply the routing table above. If ambiguous, default to claude.

Step 3: CALL THE TASK TOOL — IMMEDIATELY

Do not think about it. Do not explain what you're about to do. Call the Task tool now:

Task({
  description: "Enhance prompt for <claude|gpt>",
  subagent_type: "generalPurpose",
  prompt: "/enhance-<claude|gpt> <extracted task text>"
})

Concrete examples:

For /enhance-me Add dark mode:

Task({
  description: "Enhance prompt for claude",
  subagent_type: "generalPurpose",
  prompt: "/enhance-claude Add dark mode"
})

For /enhance-me gpt Fix the auth flow:

Task({
  description: "Enhance prompt for gpt",
  subagent_type: "generalPurpose",
  prompt: "/enhance-gpt Fix the auth flow"
})

The subagent will load the corresponding skill (enhance-claude or enhance-gpt), transform the raw task, and return the enhanced prompt as its response.

Step 4: DISPLAY the enhanced prompt — FULL TEXT, NOT A SUMMARY

After the Task subagent returns, you MUST display its entire output in your response using this exact format:

**Enhanced prompt (<Claude|GPT>):**

\```
<full enhanced prompt text from subagent — every line, no truncation>
\```

This is mandatory because:

• The user must be able to review what was generated
• It must be visible in chat history for future reference
• The main agent must see it to execute it correctly

Do NOT summarize, paraphrase, or truncate. Show the full text.

Step 5: EXECUTE the enhanced prompt

Take the enhanced prompt and execute it as your working instructions. This means:

• If it asks you to write code → write the code
• If it asks you to analyze something → do the analysis
• If it asks you to plan → produce the plan

Do NOT re-apply /enhance-me. The prompt is already enhanced. Execute it directly.

Why Subagent (Not Inline Enhancement)

• Subagent has dedicated context for enhancement best practices (XML structuring, primacy/recency, role framing, examples)
• Main agent stays focused on execution rather than meta-prompting
• Clean separation: enhancement skill evolves independently
• User sees the enhanced prompt explicitly (transparency)
• Same request, same session — no extra API cost

Protocol Compliance Check

Before ending your turn after /enhance-me triggers, verify:

• Task subagent was called (not just read about)
• Correct model skill was used (/enhance-claude or /enhance-gpt)
• Full enhanced prompt was displayed in a code fence
• Enhanced prompt was executed (work was done)

If any box is unchecked, your turn is incomplete. Go back and complete the missing step.