The famous GitHub prompt collections were written for the chatbots of 2023, and most of their prompts show it: one line of "act as a..." and a hope that the model fills in the rest. Today's models fill it in confidently and generically. The repositories are still worth reading for ideas and for the role list, and the prompt engineering guides are still worth reading in full. But a prompt that works now needs four things the old ones lack: context, constraints, an output format and a way to check the answer.
Which GitHub prompt repositories are still worth opening?
| Repository | What it is | Still useful for | Aged badly |
|---|---|---|---|
| f/awesome-chatgpt-prompts | A long list of "act as" role prompts, also published as a CSV dataset and on prompts.chat | Ideas for roles you had not thought of; a quick starting sentence | One-paragraph prompts with no context, no format and no check |
| dair-ai/Prompt-Engineering-Guide | A guide to techniques: few-shot, chain of thought, retrieval, agents | Understanding why a technique works, with papers linked | Some model-specific tricks matter less with newer models |
| anthropics/prompt-eng-interactive-tutorial | Anthropic's own step-by-step course, as notebooks | Learning to prompt Claude specifically: roles, XML tags, examples, prefilling | Little; it is maintained by the model vendor |
| openai/openai-cookbook | Worked code examples for OpenAI APIs | Developers building with APIs, structured output, evaluation | Not aimed at chat users |
The pattern is clear: lists of prompts age fast, explanations of technique age slowly. If you want a list of skills that are maintained rather than frozen, our roundup of awesome Claude skills covers the Claude side.
What is wrong with an old act-as prompt?
Take the classic shape: "I want you to act as a travel guide. I will write you my location and you will suggest a place to visit near my location." In 2023 this was a useful nudge. In 2026 it produces a fluent, generic answer, because nothing in the prompt tells the model who you are, what you have ruled out, what format you want, or how to know if the answer is good. The role is the least important part of a prompt. The context and constraints do the work.
A senior prompt engineer in your chat: system prompt design, structured output with a schema, few-shot examples, and an evaluation set so you know a change made the prompt better rather than different.
Get Noor →How do you upgrade an old prompt for Claude?
Paste the original into this and let Claude rebuild it. You will get a prompt you can save and reuse, plus the questions it needs answered to fill the gaps.
The XML-style tags are not magic, but Anthropic's own documentation recommends them for Claude because they separate instructions from material cleanly, and they make prompts easy to edit later.
Which old patterns still work?
- Giving a role. Still helps, as long as it is specific: "a tax accountant for UK sole traders" beats "an expert".
- Asking for examples first. Few-shot prompting still beats description when the format is unusual.
- Interview me first. "Ask me questions until you have what you need" is one of the best old prompts, and it works even better now.
- Step by step. Asking the model to reason before it answers still helps on multi-step problems, though newer models often do it on their own.
And the ones to drop: jailbreak-style prompts, "you have no restrictions" preambles, and anything relying on a model quirk that was patched years ago.
Need more than one specialist?
Buying Noor gets you one file. The Library subscription loads every skill in the catalog through one connector, in Claude, ChatGPT Plus and up, Claude Code, Codex or Cursor, and keeps each skill updated when we improve it. $9 a month or $59 a year, with a free trial, no card needed.
Start the free trial → Why a subscription beats buying filesShould you build your own prompt library?
Yes, but small. The prompts you reuse weekly deserve to be written properly, saved and versioned; the rest can stay in your head. A good test: if you have typed roughly the same instructions three times, save them. If a prompt is long and you use it in many chats, it is really a skill, and our explainer on agents vs skills vs prompts explains when to make the jump. For ready-made prompts with the structure above, our Claude prompts and ChatGPT prompts pages are free to copy.
Code comments, user stories, code explanations, runbooks, sprint planning and Figma-to-code handoff specs, each with context, constraints and a format, in one organised file with a table of contents.
Get the Tech & Dev library →Where do public prompt lists mislead you?
- Stars are not quality. A repository can be famous for being early, not for being good.
- No evaluation. Almost none of the list prompts were tested against a set of inputs. Treat each as an untested idea.
- Model drift. A prompt tuned for one model version can behave differently on the next. Re-test prompts you depend on.
- Licence and privacy. Check the repository licence before you republish prompts, and never paste confidential data into a prompt you found online without reading it first.
How do you install a skill in Claude?
A KissMySkills skill is a ZIP file with a SKILL.md at its root. In the Claude desktop app, open Customize → Skills, add a skill, upload the ZIP and toggle it on. Skills need code execution, which lives under Settings → Capabilities; on Team and Enterprise plans an owner controls it. Anthropic documents the steps in its Skills help article, and our step-by-step install guide has screenshots.
In ChatGPT or Gemini, open the SKILL.md in a text editor and paste it into custom instructions: a Claude Project, a custom GPT or a Gem. With a Library subscription you skip the file entirely: connect KissMySkills once and ask for the specialist by name in the chat you already use.
In summary:
Read the GitHub lists for ideas and the guides for technique, then rewrite anything you reuse with context, constraints, a format and a check. Noor ($19) does that rewrite and builds an evaluation set; the Tech & Dev Prompt Library ($9) gives you seven engineering prompts already written this way.
Noor - Prompt Engineer AI Skill
System prompts, structured output, few-shot examples and an eval set, so a prompt change is measured rather than guessed.
Or get it with every other skill in Library: $9/month, free trial, no card needed.
KissMySkills is a marketplace of 2,500+ AI skills, agents, prompt packs & free tools for Claude, ChatGPT & any AI chat.
Frequently asked questions
What is awesome-chatgpt-prompts on GitHub?
A popular open repository of role-based prompts, written in the act as format, also available as a CSV dataset and on the prompts.chat site. It is a good source of ideas but most prompts need more context to work well today.
Do ChatGPT prompts work in Claude?
Mostly yes, because they are plain text. Claude tends to respond better when instructions and material are separated, for example with XML-style tags, and when the prompt includes context, constraints and an output format.
What is the best GitHub repository for learning prompt engineering?
For technique, the Prompt Engineering Guide from DAIR.AI and Anthropic's interactive prompt engineering tutorial are the most useful. Lists of ready prompts age faster than guides that explain why a technique works.
Are act as prompts still useful?
The role alone adds little. A specific role plus your context, constraints, format and a self-check is what improves the answer.
How do I test whether a prompt is good?
Run it on five to ten real inputs, including awkward ones, and compare the answers against what you wanted. Keep those inputs and re-run them whenever you change the prompt or the model changes.
Related guides
- Awesome Claude Skills
- AI Agents vs Skills vs Prompts: What Is the Difference?
- How to Use Claude to Write Unit Tests That Catch Real Bugs
