AI Prompt Vault
Seventeen prompts documented like tools, not a wall of one-liners.
Each prompt states the job it does, which model it suits and why, the variables it takes, a worked example with the output that input actually produced — and how it fails. The failure notes are the part no prompt pack includes and the part that decides whether you keep using it.
One payment. No renewal. New prompts included.
5 are readable in full first — read them before deciding.
We already give away 40 prompts. Read this before paying.
The free library has more prompts than this does, and it always will be free. If what you need is a bigger list to paste from, go there — you will not get more of that here, you will get fewer.
What the Vault has instead is depth. Forty prompts with no worked example and no failure notes is a list. 17 prompts where you can see the output the prompt actually produced, and read the specific way it goes wrong, is a tool you can rely on. That is the entire difference, and it is worth ₹399 only if you have already hit the point where copy-paste stopped working.
What every prompt carries
All six, for all 17. Across 6 areas: Agents, RAG, Coding, Business, Data, Writing.
The job it does
A named task with a named output, not "act as an expert in X".
When to use it
And, as often, when not to.
Which model, and why
A prompt tuned for a reasoning model behaves differently on a fast one. Pretending otherwise is how people conclude prompts do not work.
Its variables
What each one is for, with an example of a good value.
A worked example
The input, and the output that input actually produced. Not a description of what it might return.
How it fails
The part no prompt pack includes. The ones that look most impressive in a demo are usually the ones that break quietly on real input.
The 17 prompts
| Prompt | Area | Access |
|---|---|---|
| Turn a vague AI idea into an architecture decisionTakes "we should use AI for X" and returns the smallest architecture that solves it, including whether AI is warranted at all. | Agents | Free |
| Choose a chunking strategy for a specific corpusGiven what your documents actually look like, returns a chunking strategy with sizes and the reason, instead of the default 1000/200 everybody copies. | RAG | Free |
| Adversarial code review that looks for the bug you have not foundReviews a diff as somebody trying to break it, not as somebody trying to be helpful. Returns concrete failure scenarios rather than style notes. | Coding | Free |
| Meeting notes into decisions, owners and open questionsSeparates what was actually decided from what was merely discussed — and refuses to invent an owner where none was named. | Business | Free |
| Build an evaluation set from real usageTurns a log of real user inputs into a graded test set, including the awkward cases you would never think to write. | Data | Free |
| Design tool schemas an agent will not misuseTakes the functions you want an agent to call and rewrites their signatures and descriptions so the model picks the right one. | Agents | In the Vault |
| Post-mortem an agent run that went wrongReads a trace and finds the step where it actually went wrong, which is rarely the step where it looks wrong. | Agents | In the Vault |
| Grade RAG answers with a rubric that survives disagreementAn LLM-as-judge prompt that grades on separate axes and is explicitly allowed to say a correct-looking answer is unsupported. | RAG | In the Vault |
| Rewrite a user question into queries retrieval can actually matchExpands a short, pronoun-laden, jargon-mismatched question into the queries your index will match, without inventing intent. | RAG | In the Vault |
| Explain unfamiliar code, including what is load-bearingExplains what code does AND which parts you must not touch, which is the half that stops you breaking production. | Coding | In the Vault |
| Find the tests that are missing, ranked by consequenceReads code and its existing tests, then names the untested paths that would actually hurt — not the ones that raise coverage. | Coding | In the Vault |
| Turn scattered incident evidence into a timeline and a causeBuilds a defensible timeline from logs, messages and dashboards, separating what is evidenced from what is inferred. | Business | In the Vault |
| Write SQL from a business question, with the assumptions statedGenerates the query and — the part that matters — the assumptions it had to make, so you find out it counted the wrong thing before you present it. | Data | In the Vault |
| Audit a dataset before you build anything on itFinds the problems that will surface later as model failures — leakage, drift, thin classes and duplicates masquerading as volume. | Data | In the Vault |
| Rewrite text so a busy reader gets it in one passCuts hedging, front-loads the point, and marks every place the original was vague rather than quietly inventing a meaning. | Writing | In the Vault |
| Release notes written for users, from commits written for engineersTurns a commit range into notes organised by what changed for the reader, and refuses to describe internal churn as a feature. | Writing | In the Vault |
| Critique a pricing page as a sceptical buyerReads your pricing page as somebody deciding whether to trust you, and names where they hesitate and leave. | Business | In the Vault |
Do not buy this if
- You want volume. There are 17. The free library at /prompts has 40 and costs nothing — if a bigger list is what you are after, that is genuinely the better page for you.
- You want prompts for social media captions, ad copy or SEO articles. These are engineering and operations prompts.
- You want something that works without being read. Each one takes a few minutes to adapt to your situation; a prompt you paste blind will disappoint you.
Questions
You have 40 free prompts. Why pay for 17?
Because they are different things. The free library is 40 prompts you copy and paste — title, category, prompt text. The Vault documents 17 of them the way you would document a tool: which model each suits and why, what each variable is for, a worked example with the real output, and the specific way each one fails. If copy-paste is working for you, stay on the free page. This is for when it has stopped working and you cannot tell why.
Can I see one first?
Yes — 5 are readable in full on the Vault page, with nothing held back: the body, the variables, the worked example and the failure notes. They are the same depth as the paid 12.
Which models do these work with?
Each prompt names the models it suits and explains the reason. Most work across Claude, GPT and Gemini; several explicitly need a reasoning model and say so, because running those on a fast model produces the generic output that makes people give up on prompting.
Do I get updates?
Yes. New prompts are added to the Vault and you keep access. One payment, no renewal.
Is it a PDF?
No — web access, so prompts can be copied cleanly and corrected when a model changes underneath them. A PDF of prompts goes stale and cannot be fixed.
Reviews
No reviews yet. The Prompt Vault is new. If you have used one of these in real work, yours would be the first — including if it did not help.
Write the first reviewRead 5 in full, then decide
Same depth as the paid 12, nothing held back. It is the only honest way to judge whether the rest are worth ₹399 to you.