💡 How to use these prompts:
Replace everything in [BRACKETS] with your specific details before sending.
Click Copy to copy any prompt to your clipboard instantly.
[SET EFFORT TO MAX BEFORE SENDING THIS PROMPT]
I have a problem that requires your deepest reasoning. Work through this step by step, explore multiple approaches, identify where you are uncertain, and arrive at the best answer you can.
Problem: [DESCRIBE YOUR HARD PROBLEM IN FULL DETAIL]
Context: [RELEVANT BACKGROUND — paste code, docs, data, or constraints]
I want you to:
1. Restate the problem in your own words to confirm understanding
2. Identify the 2-3 most promising approaches
3. Work through the best approach in full detail
4. State your conclusion with confidence level
5. Note what additional information would change your answer
Take as long as you need. This is a max-effort task.
[SET EFFORT TO LOW — this is a fast routing task, not a reasoning task]
Classify the following input into exactly one of these categories: [LIST YOUR CATEGORIES]
Rules:
- Return only the category name, nothing else
- If genuinely ambiguous, return the closest match
- Do not explain your reasoning
Input: [PASTE INPUT TO CLASSIFY]
[USE IN CLAUDE CODE OR API WITH TOOLS]
Implement the following feature in my codebase. Think through the architecture carefully before writing any code.
Feature: [DESCRIBE IN PLAIN ENGLISH WHAT IT SHOULD DO]
Repository context: [DESCRIBE YOUR STACK, KEY FILES, PATTERNS USED]
Requirements:
- [REQUIREMENT 1]
- [REQUIREMENT 2]
- [REQUIREMENT 3]
Definition of done: [HOW WE KNOW THE FEATURE IS COMPLETE]
Constraints:
- Do not modify [LIST FILES TO LEAVE ALONE]
- Match existing code style in [FILE EXAMPLE]
- All new functions must have docstrings
Work through this autonomously. After planning, implement, test, and summarise what you changed and why.
Generate the following document in full. Do not truncate, summarise, or abbreviate any section. Use the full output window available to you.
Document type: [REPORT / SPEC / PLAYBOOK / CODEBASE / OTHER]
Length target: [APPROXIMATE WORD OR PAGE COUNT]
Audience: [WHO WILL READ THIS]
Structure:
[LIST EVERY SECTION WITH A ONE-LINE DESCRIPTION OF WHAT IT COVERS]
For each section:
- Write in full, not in outline form
- Use concrete examples, not generic statements
- Include all relevant detail a reader would need without coming back to ask
Source material to draw from: [PASTE REFERENCE MATERIAL]
Begin with Section 1 and continue until the document is complete.
I am testing how Claude Opus 5's automatic fallback handles edge cases in my application. I want to understand which request types route to a fallback model vs which Opus 5 handles directly.
Please respond to the following requests in sequence and indicate after each one whether you handled it directly or whether it seems like a case where your fallback would normally trigger:
Request 1: [ROUTINE TASK]
Request 2: [MILDLY SENSITIVE TOPIC RELEVANT TO YOUR APP]
Request 3: [BORDERLINE CASE FOR YOUR USE CASE]
Request 4: [CLEARLY SENSITIVE CASE]
For each: answer the request if you can, then add one line: "Handled directly" or "This type would normally trigger fallback" with a brief reason.
This helps me calibrate my application's expectation of where fallback routing applies.
[USE WITH API — mid-conversation tool changes beta feature]
You are an autonomous research and action agent. I will give you a research objective. Work through it using the tools available to you. When the research is complete, I will update the tools available to you mid-conversation without invalidating the cache — at that point use the new tools to take action on what you found.
Research objective: [DESCRIBE WHAT YOU NEED TO FIND OUT]
Phase 1 — Research (use search and read tools):
1. Research the objective thoroughly
2. Produce a structured findings report
3. Identify the 3 most important actions to take based on findings
4. Signal you are ready for Phase 2 with: "RESEARCH COMPLETE — READY FOR ACTION TOOLS"
I will then add action tools (write, send, update) and you will proceed to Phase 2 without cache invalidation.