The Prompt Index
описание
AI news, AI ChatGPT prompts, Claude prompts, Gemini prompts, Midjourney and other AI prompts, prompt techniques and so much more. ChatGPT, OpenAI, Midjourney, AI art, Prompt Engineering. Artificial Intelligence Link: https://linktr.ee/thepromptindex
7 648
подписчиков
Охват к подписчикам
6,3%
ERR
Реакции к просмотрам
0,35%
76 на 50 постов
Пересылки к просмотрам
0,47%
102
Постов в день
3,6
всего 48
Где отзываются чаще
доля реакций к просмотрам- 15 авг.без подписи4,17%
- 15 авг.без подписи3,57%
- 15 авг.без подписи3,33%
- 15 авг.без подписи3,23%
- 15 авг.без подписи3,13%
- 16 авг.Claude Code: Reviewer-Spawns-Verifiers Review Pattern Aimed At: Engineering teams using Claude Code who want review findings independently checked instead of self-graded, without cluttering the main conversation with the back-and-forth. Benefits: Also benefits solo developers running Claude Code unattended, QA engineers who want a second opinion built into the review step, and teams standardising subagent architecture across projects. Compatible Models: Claude Category: Coding The Prompt: You are the code-review coordinator for this change set. 1) Read the diff and produce a numbered list of findings (bug, security, style, or performance). 2) For EACH finding, spawn a separate verifier subagent whose only job is to confirm or refute that specific finding against the actual code — never verify your own findings. 3) Wait for every verifier to report back. 4) Discard any finding a verifier refutes. 5) Return to the main conversation ONLY a short, prioritised summary of the confirmed findings — keep the reviewer/verifier exchange out of the main thread. What this prompt does It stops Claude Code from grading its own homework: a reviewer subagent lists findings, then fans out a dedicated verifier per finding to independently confirm or refute it, and only the confirmed, prioritised list surfaces in your main conversation. Worked example Drop this into a project's CLAUDE.md or a custom /review-pr command. A reviewer flags five issues in a pull request, spawns five verifiers in parallel, two findings get refuted and dropped, and your main conversation only ever sees the three confirmed issues with a one-line fix suggestion each. Editor's note: This relies on Claude Code's nested subagent depth, currently capped at five levels below the main conversation — a subagent at depth five doesn't get the Agent tool, so it can't spawn further agents. Keep the pattern to one reviewer layer plus one verifier layer rather than chaining deeper. Source Link: https://www.xda-developers.com/set-up-claude-way-anthropic-now-recommends-sub-agents/1,10%
- 9 авг.And just so I don't get messages asking about the prompt, here is that prompt in that screenshot Reasoning Pathway Generator You are a Reasoning Architecture Designer. Task: [[task]] Generate multiple independent reasoning pathways for solving the task. Create: Pathway 1, Direct Analysis Pathway 2, First Principles Pathway 3, Constraint Analysis Pathway 4, Alternative Hypotheses Pathway 5, Adversarial Analysis For each pathway provide: Approach Key questions Decision points Evidence requirements Expected conclusion Do not expose hidden chain of thought or private reasoning. Provide concise reasoning summaries and decision criteria instead.0,82%
- 17 авг.You wanna hear something funny? This was not even made by a chatbot. It was made by something for a completely different purpose. [SYSTEM] Role: You are an Expert Prompt Engineer with mastery in structural prompt design and adversarial hardening. Your purpose is to architect high-fidelity meta-prompts that can be used to control other AI models. Core Behaviors: * Step Sequencing: You must follow a 4-phase "gauntlet" process for every request: Analysis, Architecture, Hardening, and Validation. * Chain of Thought: Think through the logic of the prompt step-by-step, making your reasoning explicit before showing the final result. * Knowledge Boundary: If a requested task is outside the model’s safe operating parameters, you must flag it and suggest a safe alternative. Phase 1: Diagnostic & Discovery Before building, perform a Prompt Anatomy on the user’s request. * Identify the Role, Task, and Audience. * Ask 3 Clarifying Questions to resolve any ambiguity in the user’s intent. * Identify potential Blind Spots or unstated assumptions. Phase 2: Structural Architecture Construct the draft meta-prompt using the following core blocks: * Persona Block: Define the identity, expertise, and behavioral rules. * Instruction Block: State directives in plain, imperative language with priority levels. * Context Injector: Layer in domain knowledge and success metrics. * Variable Extractor: Identify and define all elements that should be dynamic variables (e.g., [[variable_name]]). Phase 3: The Hardening (The Gauntlet) Subject the draft to a Red Teaming Self-Critique. * Anti-Pattern Detector: Scan for vague verbs, missing constraints, or conflicting instructions. * Advanced Bias Checker: Apply the Bias Bingo framework to ensure the prompt is neutral and inclusive. * Scope Lock: Add explicit "MUST NOT" instructions to prevent model drift or hallucinations. Phase 4: Final Output & Schema Provide the final result in a Layered Output format: 1. Level 1 (TL;DR): A one-sentence summary of what the prompt does. 2. Level 2 (The Prompt): The full, production-ready system message wrapped in clear delimiters. 3. Level 3 (Execution Guide): A Step-by-Step checklist for the user to implement the prompt effectively. [/ SYSTEM] How this Gauntlet Works: 1. Decomposition: It uses the Task Decomposition component to break complex user requests into atomic, executable parts. 2. Refinement: It utilizes Recursive Prompting logic, requiring the AI to critique and improve its own draft before you ever see it. 3. Variable Management: It incorporates the Variable Extractor to ensure the resulting meta-prompt is reusable across different datasets. 4. Safety: By embedding the Prompt Hardener and Scope Lock, it ensures the generated prompts are robust against adversarial misuse.0,75%
- 14 авг.I need 5 active play testers to test a Gemini prompt that converts a video into a floor plan. "Active" means I can message you and receive a response without waiting two days.0,71%
- 9 авг.That's why prompt engineering guides you to treat the AI like a new intern. Would you expect someone you just hired to know exactly what you want without talking to them, explaining your goals, and ensuring they understand? That is exactly how you should approach the chatbot. Talk to it, clarify what it needs to know, and build that understanding together. So I'm going to say again: stop trying to get 100% quality in one shot in the first response. Regardless of what plan you're on, that's not going to work. if you want better responses, you're goanna have to train to get better responses. You can't just get it. in one shot. Regardless of what version of opus is out there what version of luna or SOL it doesn't work that you have to talk to it. That's literally the joy of it: you're able to customize it and tell it how you want it, instead of letting it just assume.0,66%
- 15 авг.Hey everyone. If you want to participate in free giveaways, free prompt requests, and more, you need to be contactable. Please make sure you have a username available and that your DMs are open. I need to be able to contact you directly without having to request access. If your settings only allow specific people to contact you, please turn those restrictions off. Otherwise, it can be very difficult for me to reach you.0,66%
- 9 авг.The important part is that ARCHITECT is designed to create prompt systems, not just prompt text. It looks underneath the wording to identify the mechanism that makes a prompt useful, then turns that mechanism into something you can reuse, modify, version, and combine with other prompts. Best use cases ARCHITECT is particularly useful for: Discovering advanced prompting techniques Building a personal prompt library Designing reusable prompt templates Finding overlooked combinations of prompting techniques Creating prompts that can survive model updates Turning one prompting concept into multiple architectures Developing prompt engineering frameworks Building modular prompt systems Creating prompts for different stages of an AI workflow Designing prompts that can later be chained or combined Think of it as: Idea → Technique Discovery → Architecture → Prompt Template → Testing → Versioning → Prompt Library0,56%