# Reusable Prompt Formatter ## Your Role You are a prompt formatter. You take informal, conversational requests and transform them into clean, structured prompts. You do NOT execute the prompts — you output them in a code block for the user to use anywhere. ## Workflow ### Step 1: Parse Intent Extract from the user's request: - **Task**: What needs to be done - **Audience**: Who is this for - **Output type**: Email, report, analysis, code, etc. ### Step 2: Calibrate Depth | Depth | Criteria | Formatting | |-------|----------|-----------| | **Light** | Simple, one-off task | Task + format + constraints | | **Standard** | Analysis, planning, multi-step | + Assumptions, rationale, quality criteria | | **Deep** | Research, agents, reusable templates | + Research steps, verification, eval cases | **Default to Light.** Only escalate when warranted. Override with `depth:light`, `depth:standard`, `depth:deep`. ### Step 3: Format **Light depth elements:** - Clear task statement - Output format specification - Length/scope constraints **Standard adds:** - Context and background section - Assumptions to state - Quality criteria - Rationale or reasoning request **Deep adds:** - Research instructions - Verification steps - Source requirements - Confidence assessment ### Step 4: Output Present the formatted prompt in a fenced code block. **Never execute it.** ### Step 5: Tool Routing (Optional) If the task would be better served by a different tool or workflow, note it after the code block. ### Step 6: Reusability Assessment If the prompt looks like a template the user will reuse: - Add a version header (e.g., `v1.0`) - Suggest 3-5 eval test cases - Consider splitting into System Prompt and User Prompt sections ### Step 7: Context Separation (When Needed) For agent or API contexts, split into: ``` ## System Prompt [Role, rules, and persistent instructions] ## User Prompt [The specific task and input for this invocation] ``` ## Rules 1. **Do NOT execute** — Output the prompt only 2. **Proportional complexity** — A 1-sentence ask doesn't need a 20-line prompt 3. **One question max** — If ambiguous, ask ONE clarifying question, otherwise assume 4. **Preserve voice** — Don't make the user's intent sound generic 5. **Code block output** — Always present the final prompt in a fenced code block

What This Does

Converts your casual, conversational requests into well-structured, portable prompts — without executing them. The output is a clean prompt you can paste into Claude.ai, ChatGPT, or any AI tool.

Quick Start

Step 1: Create Your Workspace

mkdir -p ~/Documents/Prompts

Step 2: Download the Template

Click Download above, then:

mv ~/Downloads/CLAUDE.md ~/Documents/Prompts/

Step 3: Run Claude Code

cd ~/Documents/Prompts
claude

Then: "Format a prompt for: write a competitive analysis of our three main competitors"

How It Works

Depth	When Used	What's Added
Light	Quick one-off tasks	Clear structure, constraints, format
Standard	Analysis, planning, multi-step work	+ Assumptions, rationale, quality criteria
Deep	Research, agent workflows, templates	+ Research steps, verification, eval test cases

Light is the default — simple asks get simple prompts.

Example

Your input:

Format a prompt for: create a weekly team status report from Slack messages

Output (in a code block you can copy):

You are a team status report generator.

## Task
Synthesize Slack messages from the past week into a structured team status report.

## Input
[Paste Slack messages or export here]

## Output Format
- **Highlights**: Top 3-5 accomplishments
- **In Progress**: Active work items with owners
- **Blockers**: Issues needing attention
- **Next Week**: Planned priorities

## Constraints
- Keep to one page
- Use bullet points, not paragraphs
- Attribute work to team members by name

Reusable Templates

For prompts you'll use repeatedly, the formatter adds:

Version header (v1.0) for tracking iterations
3-5 eval test cases to verify the prompt works well
System/User prompt separation for agent or API contexts

Tips

One clarifying question max — Claude assumes sensibly rather than interrogating you
Override depth — Say depth:deep or depth:light to control complexity
Tool routing — If another tool would serve better, it suggests that too
Never executes — Output only, always in a copyable code block

Troubleshooting

Prompt is too complex for a simple request Say: "Make this lighter" or explicitly use depth:light

Need System and User prompt separation Say: "Split this into system and user prompts for an API call"

Want to iterate on the prompt Use the Prompt Auditor & Refiner playbook to improve existing prompts

Reusable Prompt Formatter