{
 "version": "https://jsonfeed.org/version/1.1",
 "title": "Prompt Verse blog",
 "home_page_url": "https://prompt-verse.io/blog/",
 "feed_url": "https://prompt-verse.io/blog/feed.json",
 "description": "Practical posts on versioning, testing and sharing prompts, from the team building Prompt Verse.",
 "language": "en",
 "items": [
  {
   "id": "https://prompt-verse.io/blog/your-prompts-are-already-software/",
   "url": "https://prompt-verse.io/blog/your-prompts-are-already-software/",
   "title": "Your prompts are already software, so give them version control",
   "summary": "A prompt that decides what your product says is code in everything but name. Four things software teams do that prompt owners usually skip.",
   "date_published": "2026-10-17T09:00:00Z",
   "authors": [
    {
     "name": "The Prompt Verse team"
    }
   ],
   "tags": [
    "versioning",
    "product"
   ],
   "content_text": "If a sentence you wrote decides what your product tells a customer, it is code. It has inputs, it has behaviour, and it can be wrong. The only thing that makes it feel different is that it is written in English, so we keep it where we keep English: in documents, chat threads and comments.\nThat works until the first time it does not. A model gets updated. Someone tidies the shared doc. A good wording from March is buried in a conversation nobody can find. Then the question is always the same: what changed, and was it better?\nSoftware teams solved this long ago. Here are the four habits worth stealing.\n1. Keep one source of truth\nA prompt that exists in five places has no current version. Pick one place, and make everything else read from it. In practice that means your code, your agents and your colleagues all ask the library for the prompt instead of carrying a pasted copy.\nThe cost of skipping this is invisible until it bites: two people use two wordings, and the product behaves differently depending on who answered.\n2. Commit with a reason\nA commit is a saved state plus a sentence. The sentence is the valuable part. “Cut to six sentences, no promises” tells the next person what you tried and why. “Update” tells them nothing.\nYou do not need a clever process. You need the habit of saving before you change something, and of writing one line afterwards.\n3. Change on a branch\nTry new wording where it cannot hurt. A branch is a copy of the history that you can edit freely while the version in use stays put. If the experiment works, you merge it. If not, you throw the branch away and nothing happened.\nMerging matters more than it sounds. Two people will, sooner or later, edit the same prompt in different directions. A merge that shows you both sides and asks you to choose is better than a shared document where the last writer wins.\n4. Test before you ship\nThis is the habit most prompt owners skip, because testing text feels vague. It does not have to be. Write down five inputs that matter, including the awkward ones, and what a good answer contains. Run the old and the new wording on them and compare. That is an eval, and even a crude one beats a feeling.\nWhat this does not need\nNone of this needs a built-in AI. Versioning, branching and organising are file-and-history problems. The running can stay where it already is, in your own assistant.\nThat is the idea behind Prompt Verse: a workspace that gives prompts, agents, workflows, MCP tools and skills a home, a history and a way to be checked, and leaves the running to the AI you already use.\nIf you take only one thing from this post, take this: before you change a prompt that matters, save the old one with a sentence. Everything else follows from that."
  },
  {
   "id": "https://prompt-verse.io/blog/write-prompts-with-variables/",
   "url": "https://prompt-verse.io/blog/write-prompts-with-variables/",
   "title": "How to write a prompt with variables that does not break",
   "summary": "Turn a prompt you retype into a template. How to choose variable names, what to keep out of them, and how to find the missing ones before your AI does.",
   "date_published": "2026-10-17T09:00:00Z",
   "authors": [
    {
     "name": "The Prompt Verse team"
    }
   ],
   "tags": [
    "how-to",
    "variables"
   ],
   "content_text": "If you retype the same prompt and change two words, you have a template waiting to be written. A template has a fixed part and a few variables. In Prompt Verse a variable is any name in double braces, like {{customer_name}}.\nHere is a way to do it that holds up.\nStart from a prompt that already worked\nDo not design a template from nothing. Take a prompt you used last week and highlight the words that changed between uses. Those are your variables. Everything else is the template.\nReply to this customer in a friendly, short way. Their name is Sanne. Their order is A-1043 and it is two days late. They wrote: \"Where is my order?\"\nReply to this customer in a friendly, short way.\nTheir name is {{customer_name}}.\nTheir order is {{order}} and it is {{days_late}} days late.\nThey wrote: \"{{message}}\"\nName variables for what they hold\n{{message}} is better than {{text1}}. {{days_late}} is better than {{n}}. You, a colleague and your AI will all read these names, and the name is the only documentation a variable gets.\nUse letters, digits and underscores. Keep names short and singular.\nKeep rules out of variables\nA variable is for data that changes per use. A rule belongs in the prompt itself, preferably as a wrapper. If you notice a variable that is almost always the same value, it is not a variable, it is part of the template.\nPut the awkward part last\nModels handle long free text best when the instruction comes first and the unknown data comes last. Variables that hold long text, like {{message}}, belong at the end of their section.\nFind the missing ones before your AI does\nWhen you render a prompt in Prompt Verse, the result lists any variable you did not fill. That is the cheapest test there is: render with an empty set of values and read the list. It tells you exactly what the prompt expects.\nOver MCP the same thing works for your assistant. It asks for the rendered prompt, sees which variables are missing and asks you for those values instead of guessing.\nCheck how it looks in other formats\nThe same template can be rendered as Markdown, text, XML, JSON, YAML, HTML, CSV, TOML, JSONL, chat messages or Prompty. If you plan to send the prompt to an API that takes a message list, render it as messages once and read the output. Structure mistakes show up immediately.\nA short checklist\nEvery variable has a name that says what it holds.\nNo variable holds a rule.\nThe long free-text variable comes last in its section.\nA render with no values lists exactly the variables you expect.\nThe template is committed with a message.\nThat last line is the one people forget. Templates get edited more often than one-off prompts, so save the version that worked before you tidy it."
  },
  {
   "id": "https://prompt-verse.io/blog/test-a-prompt-with-five-cases/",
   "url": "https://prompt-verse.io/blog/test-a-prompt-with-five-cases/",
   "title": "Test a prompt with five cases and no tooling",
   "summary": "A practical way to check whether a changed prompt is better: pick five inputs, write what a good answer contains, score both versions and keep the winner.",
   "date_published": "2026-10-17T09:00:00Z",
   "authors": [
    {
     "name": "The Prompt Verse team"
    }
   ],
   "tags": [
    "evals",
    "how-to"
   ],
   "content_text": "“Is the new wording better?” is the most common question about prompts and the least often answered with evidence. You do not need a test framework to answer it. You need five cases and an honest hour.\nStep 1: pick five inputs that matter\nDo not pick five typical inputs. Pick the ones that have gone wrong, or could:\nThe angry message.\nThe message with a missing detail (no order number).\nThe message that is not a question at all.\nThe longest message you have seen.\nThe one that nearly caused an incident.\nIf you cannot find five, find three. The point is to cover failure, not the middle.\nStep 2: write what a good answer contains\nFor each input, write one line about the answer. Not the whole answer, just a check you can apply:\n“Says what happens next and when.”\n“Does not promise a refund.”\n“Asks for the order number.”\n“Stays under six sentences.”\nWhere you can, make the check mechanical. “Contains the words ‘order number’” is a check a machine can apply. “Sounds kind” needs a judge, either a person or a model.\nStep 3: choose a scoring method\nPrompt Verse offers five: exact, contains, structural, judge and hybrid. Use the simplest that fits.\n`contains` for “must mention” checks.\n`structural` when the answer must be valid JSON with certain keys.\n`judge` when quality is the point and you accept that a model does the grading.\n`hybrid` when you want a rule and a judgement together.\nStep 4: run both versions on the same cases\nThis is the part that matters. Run the old wording and the new wording on the same five inputs and score each answer from 0 to 1. Average them. Set a threshold you are willing to defend, 0.8 is the default in Prompt Verse.\nIn the pre-beta Prompt Verse runs no model, so your own AI does this step: it reads the saved cases, runs each version and scores the answers in the conversation.\nStep 5: keep the winner and say why\nCommit the version that scored higher with a message that includes the numbers: “Shorter reply. Mean 0.90, was 0.78 on five cases.” Six months from now that sentence is worth more than the diff.\nWhat this will and will not tell you\nFive cases will not prove a prompt is good. They will catch the regressions you already know about, and they will make you articulate what “good” means, which is half the value. When a new failure appears in the wild, add it as case six. That is how a test set grows, and it is how prompts stop being fragile."
  },
  {
   "id": "https://prompt-verse.io/blog/connect-your-ai-over-mcp/",
   "url": "https://prompt-verse.io/blog/connect-your-ai-over-mcp/",
   "title": "Connect Claude or ChatGPT to your prompt library over MCP",
   "summary": "What MCP is, why it beats copy and paste, and the steps to let your AI search, render and improve prompts in Prompt Verse, in about five minutes.",
   "date_published": "2026-10-17T09:00:00Z",
   "authors": [
    {
     "name": "The Prompt Verse team"
    }
   ],
   "tags": [
    "mcp",
    "how-to"
   ],
   "content_text": "Copying a prompt out of a library and pasting it into a chat works. It also does not scale: you copy the wrong version, you forget a variable, you lose the improvement you made in the chat. A better arrangement is for your AI to ask the library directly.\nThe open standard for that is MCP, the Model Context Protocol. An MCP server offers tools, and an AI assistant that supports MCP can call them. Prompt Verse is an MCP server for your prompts, agents, workflows, tools and skills.\nWhat your AI can do once connected\nSearch everything you own by title, tag or text.\nRead a prompt, or ask for it **rendered**: pieces resolved, wrappers applied, variables filled, in the format you name.\nUpdate a prompt, commit a version, tag it.\nRead the cases of an eval, so it can test a prompt for you.\nIt cannot make Prompt Verse run a model. In the pre-beta that is switched off on purpose, and the error message tells your AI to render the prompt and answer it itself.\nConnect in five minutes\nOn its own, Prompt Verse has an address of its own that ends in /mcp. You find yours in the app under Connect.\nIn Claude: open Settings, then Connectors, then Add custom connector. Paste your Prompt Verse address, save and sign in when asked.\nIn ChatGPT: open Settings, then Connectors, add a remote MCP server, paste the address and sign in.\nOther assistants and scripts: make a key under Keys in the Connect page, point the client at your address and send the key as a bearer token. The key is shown once, so copy it then.\nIf you use Geenius OS, there is a shorter route: install Prompt Verse in your workspace and your AI gets a tool called geenius_promptverse through the one connector it already has.\nYour first ask\nUse Prompt Verse. Find my prompts about support replies, open the best one, render it as chat messages for this message: \"Where is my order?\", and tell me which variables are still missing.\nIf the connection works, your AI will search, open the prompt, render it and tell you what is missing.\nTwo habits worth having\n**Ask for the render, not the raw prompt.** The rendered text is what you actually want to run, and the render tells you what is missing.\n**Let it suggest, then commit yourself.** Ask your AI to propose an improvement. Read it, then commit it with a message. The history stays yours.\nA note on keys\nNever paste a key into a chat. Use a key only in the configuration of a client that cannot sign in, name it after what uses it, and revoke it when you stop. If you can sign in instead of using a key, do that."
  },
  {
   "id": "https://prompt-verse.io/blog/why-prompt-verse-runs-no-model/",
   "url": "https://prompt-verse.io/blog/why-prompt-verse-runs-no-model/",
   "title": "Why Prompt Verse runs no model",
   "summary": "The reasoning behind a prompt workspace without a built-in AI: your choice of model stays open, there are no credits to meter, and the storage stays honest.",
   "date_published": "2026-10-17T09:00:00Z",
   "authors": [
    {
     "name": "The Prompt Verse team"
    }
   ],
   "tags": [
    "product",
    "mcp"
   ],
   "content_text": "Most tools in this space start by putting a model inside the product and selling credits. Prompt Verse did the opposite in the pre-beta, and it is worth explaining why.\nThe thing people actually have a problem with\nPeople do not struggle to get a model to answer. They already have Claude, ChatGPT or Grok, and they pay for it. They struggle to keep, change and trust the prompts they use with those models. That is a storage, history and testing problem, and it does not need a model to solve.\nSo the first version solves that problem and leaves the answering to the AI you already use.\nWhat it saves you\n**No second subscription for the same job.** Your AI runs the prompt on your own plan. There are no tokens or credits to buy from us.\n**No lock-in to one model.** A prompt is rendered text. It works with whichever assistant you use next year.\n**No surprise costs.** Rendering, versioning and sharing are plain code. The plans are priced for the workspace, not for compute.\nWhat it costs you\nYou have to connect your AI. Through MCP that is a one-time step of a few minutes, but it is a step. And things that need a model inside the product, like running an eval and keeping the scored run, are switched off in the pre-beta. Your AI can score cases for you in a conversation, but the result does not land in the app by itself.\nWe think that is the right trade for now. We would rather be exact about what works than ship a half-built runner to look complete.\nWhat changes later\nThe Premium and Max plans carry the AI features of the app, running prompts, agents, workflows and evals inside Prompt Verse. The code exists and is switched off during the pre-beta. When that changes, the changelog will say so.\nThe short version\nIf you take only the storage and history parts of Prompt Verse and nothing else, you still have something useful: a library, versions, evals and sharing. The model is a guest, not the landlord."
  },
  {
   "id": "https://prompt-verse.io/blog/folder-git-or-prompt-verse/",
   "url": "https://prompt-verse.io/blog/folder-git-or-prompt-verse/",
   "title": "A folder of text files, Git, or Prompt Verse?",
   "summary": "An honest comparison of three ways to keep prompts: a folder or notes app, a Git repository, and a prompt workspace. What each is good at, and when to pick which.",
   "date_published": "2026-10-17T09:00:00Z",
   "authors": [
    {
     "name": "The Prompt Verse team"
    }
   ],
   "tags": [
    "comparison",
    "versioning"
   ],
   "content_text": "You do not need a product to keep prompts. Plenty of people do fine with less. This is an honest look at three approaches and when each is the right one.\nA folder of text files, or a notes app\nGood at: being there. Zero setup, zero learning, works with every tool you own.\nWhere it hurts: history and search. You usually have no record of what changed, and finding a prompt means remembering its name. Sharing means sending a file. Two people editing the same file is a race.\nPick it when: you have a handful of prompts, you work alone, and nothing breaks if one changes.\nA Git repository\nGood at: history. Commits, branches and merges are exactly the right tools, and your team already knows them. Prompts sit next to the code that uses them and go through the same review.\nWhere it hurts: everyone who touches a prompt needs to be comfortable with Git. Non-developers find it hard. Prompts are not just text files: they have variables, they render into several formats, and you want to search across kinds of assets, none of which Git knows about. And an AI assistant cannot ask a repository for a rendered prompt.\nPick it when: your prompts live inside a codebase, your team is all developers, and your prompts are few and mostly static.\nA prompt workspace like Prompt Verse\nGood at: treating prompts as first-class things. Typed pieces, variables, rendering in eleven formats, commits and branches without Git, evals, collections, sharing by link, and a door for your AI over MCP. The same model covers agents, workflows, tools and skills.\nWhere it hurts: it is another place to keep things, and it is in pre-beta. You need an account and a place on the list. There are no client libraries yet, so using it from code means the HTTP API or MCP. It runs no model, so you still bring your own AI.\nPick it when: several people, several kinds of asset or several AIs touch the same prompts, you want to test changes, or non-developers need to edit.\nA simple rule of thumb\nOne person, a few prompts: a folder.\nA codebase with a few static prompts: Git.\nMany prompts, many people, many tools, a need to test: a workspace.\nYou can also combine them. Nothing stops you from keeping the final tagged version of a prompt in your repository and the workshop in Prompt Verse.\nWhat we would not claim\nWe would not claim Prompt Verse is better than Git at what Git does. It is not a replacement for it. It is a different tool for a different shape of problem, where the thing you are versioning is not a source file but a prompt with variables, a format and an audience."
  }
 ]
}