Home / All prompts / meta

Meta-Prompting prompts

Meta-prompting for designing, diagnosing, and improving prompts themselves. 28 prompts in this category, 28 readable for free.

• Diagnose and improve a prompt

  Here is my prompt: {{prompt}}. What I want: {{desired}}. What I got: {{actual output}}. Diagnose: (1) which elements are missing — role, task, input,

• Turn a vague request into a clear task

  I have a vague idea: {{vague request}}. Turn it into an executable task definition: (1) state in one sentence what will be produced — a concrete deliv

• Make the model ask before answering

  I need to do {{task}} but the information is incomplete. Do not answer yet. First: (1) list what you must know to do this well, at most six items; (2)

• Design few-shot examples

  I need consistent output in a specific style or format: {{requirement}}. Design three few-shot examples: (1) each with input and ideal output, where t

• Two-pass self-critique

  Complete this task: {{task}}. Deliver in two passes: first your draft; then switch to a strict reviewer and raise five specific criticisms of your own

• Enforce a strict output format

  Output strictly in this format with no extra text: {{format definition, such as a JSON schema or column names}}. Rules: (1) output only the structured

• Set an effective role

  I need a role setup for this task: {{task}}. Write it so that: (1) it is not a hollow 'you are a senior expert' but three concrete traits — what this

• Process long documents in stages

  I need to process a long document: {{description}}. Design a staged workflow to avoid omissions from one-shot processing: (1) stage one extracts only

• Require confidence labeling

  Answer this question: {{question}}. Requirements: (1) label each conclusion — [certain] with reliable support, [likely] with indirect support, [guess]

• Multi-role red-team debate

  Stress-test this proposal: {{proposal}}. Run a three-way debate in four rounds: (1) the advocate ({{role A}}) makes the strongest case; (2) the oppone

• Ask for method, not answers

  I am working on this problem: {{problem}}. Act as a coach and do not hand over the answer. Rules: (1) first gauge my level by asking one diagnostic qu

• Turn a prompt into a reusable template

  Turn this prompt into a reusable template: {{prompt}}. Requirements: (1) extract variable parts as {{variable_name}} placeholders that are self-explan

• Prevent fabrication with sourcing rules

  Follow these rules strictly: (1) state only what you can source specifically — which document, which institution, what date — never a bare 'studies sh

• Iterative follow-up questioning

  Continuing from your last answer. Your conclusion was: {{prior conclusion}}. (1) List the three premises that conclusion depends on; (2) I question th

• Naming and versioning a prompt library

  I am organizing my prompt library — about {{count}} entries, currently messy. Design a management scheme: (1) a naming convention that reveals scenari

• Define structured output with a JSON schema

  You are a prompt-engineering expert. Make the model output structured data: (1) define fields, types and required keys; (2) give a JSON Schema example

• Break a task into steps and orchestrate agents

  You are an AI-workflow designer. Decompose a complex task: (1) map the input-to-intermediate-to-final chain; (2) define each step's role, inputs, outp

• A/B test two prompts

  You are a prompt-evaluation expert. Compare two prompts: (1) set criteria (accuracy, format, style, length); (2) a fixed test-input set; (3) a scoring

• Use chain-of-thought prompting

  You are a prompt-engineering expert. Use chain-of-thought: (1) ask the model to think step by step and show reasoning; (2) when CoT helps and when it

• Use self-consistency

  You are a prompt-engineering expert. Explain self-consistency: (1) sample multiple answers and take the majority; (2) how it improves complex reasonin

• Use tree-of-thoughts

  You are a prompt-engineering expert. Explain tree-of-thoughts: (1) explore, evaluate and backtrack across branches; (2) how it differs from chain-of-t

• Use ReAct (reason + act)

  You are an AI-workflow expert. Explain ReAct: (1) alternate reasoning and acting (tool calls); (2) a thought-action-observation loop; (3) how it works

• Manage context window

  You are a prompt-engineering expert. Manage context: (1) context windows and forgetting; (2) compress via summary, chunking and retrieval; (3) a strat

• Design a system prompt

  You are a prompt-engineering expert. Design a system prompt: (1) role, goal and boundaries; (2) behavior rules and prohibitions; (3) output format and

• Design a multi-turn dialogue strategy

  You are a dialogue-system expert. Design a multi-turn strategy: (1) what to collect each turn; (2) follow-ups, clarifications and fallbacks; (3) handl

• Tune temperature and parameters

  You are a prompt-engineering expert. Tune parameters: (1) explain temperature, top_p and others; (2) recommended values by task (creative vs precise);

• Defend against prompt injection

  You are an LLM-security expert. Defend against prompt injection: (1) how injection works and common tricks; (2) set boundaries and priority in the sys

• Optimize prompt cost

  You are a prompt-engineering expert. Optimize cost: (1) trim redundant prompt text; (2) smaller models or tiered strategy; (3) cache and reuse; (4) ba

Other categories

Get the full library

Locked prompts are part of the full version — 969 prompts, one-time payment.

See pricing