How to run a prompt library with this
How to write an entry worth keeping, and how a team shares one library without a server.
- What it is
- Using a prompt
- Writing an entry
- Sharing with a team
- Keeping it useful
- Notes per target
- Where the data lives
What it is
Prompt Library is a single HTML file. It keeps a list of prompts in your browser, lets you fill in the variable parts and copy the result, and reads and writes a JSON file so a team can pass the library around. No account, nothing to install. Open the file (or the hosted link) and it works.
It ships with twelve example prompts, three for each of the four places a prompt goes in Articulate 360: a Rise 360 code block, a Storyline 360 JavaScript trigger, the AI Assistant for images, and the AI Assistant for text. Edit them or delete them; until then they double as the pattern for how an entry should read.
Using a prompt
- Find it. Search matches titles, prompt text, tags and contributor names. The Goes into list narrows to one target; tags narrow further (selecting several tags shows prompts that carry any of them).
- Open it by clicking the row. Each row has a running number, so "use 007" means the same thing to everyone on the team.
- Fill the highlighted parts. Anything written as
like thisis a field you type into. If the same field appears twice in a prompt, typing in one fills both. - Copy prompt. The filled text goes to the clipboard. A field you left blank stays as
{{its_name}}in the copy, so you will notice it before the AI does. - Paste it where it goes. Read What you get to know what a good result looks like, and What to change for the things people trip on with this one.
Writing an entry
Prompt libraries usually die the same way: a wall of pasted prompts, and six weeks on nobody remembers which one worked. The fields below exist to stop that.
| Field | What it is for |
|---|---|
| Title | What the prompt produces, in five words or fewer. "Flip-card grid", not "Prompt for cards v2 final". |
| Goes into | The one place the output is pasted. If a prompt is used in two places, make two entries. |
| Prompt | The exact text, with every part the user must change marked as placeholder. Include the constraints that took you three attempts to learn (no external files, report the iframe height). |
| What you get | One or two lines describing the shape of a good result, so a colleague can tell at a glance whether their output is right. |
| What to change | The advice you would give over someone's shoulder: what to feed it, and what to check before shipping. This is the field people skip, and the one that makes a prompt reusable. |
| Tags | Lower case, one idea each. Agree a short list as a team (see below). |
| Added by | Your name, so the team knows who to ask. Remembered between entries. |
Placeholders
Wrap the variable part in double curly brackets: {{topic}}, {{source_text}}. Use underscores for spaces; they display as spaces. Name the placeholder for what goes in it, not what it is: {{reading_age}} beats {{level}}. Big pasted inputs (a policy document, a list of steps) work fine as a placeholder; the field grows as you type.
Before you add it
- Run it once yourself, today, and put the result in What you get.
- Search first. If a near-identical prompt exists, edit that one rather than adding a rival.
- Strip anything specific to one project (client names, exact colours) into a placeholder, unless the prompt only makes sense for that project, in which case tag it with the project.
Sharing with a team
Each person's copy of the library lives in their own browser. The team shares it as a JSON file, and the tool merges files so nobody overwrites anyone else's work. The workflow that holds up in practice:
- One owner. Someone keeps the master file. It lives where the team already works, a Teams channel or a SharePoint folder. Call it
prompt-library-master.json. - Contributors export. When you have added or improved prompts, click Export JSON and drop the file in the shared folder, or send it to the owner. Exporting includes your whole library; that is fine, the merge sorts it out.
- The owner imports and prunes. Import each contributor file into the master copy, delete anything duplicated or dead, then export the master again.
- Everyone imports the master. Monthly is enough. Import never deletes your own prompts, so it is safe to do at any time.
How the merge decides
Every entry has a hidden ID assigned when it is created. On import, the tool compares IDs:
- An ID you do not have is added.
- An ID you do have, where the incoming copy was edited more recently, is offered as a replacement.
- An ID you have, where your copy is the same age or newer, is left alone.
If anything would be replaced, the tool stops and lists it before changing anything: each prompt's title, the date and author of your copy, and the date and author of theirs. Everything is ticked by default; untick a prompt to keep your version. Cancel import leaves your library exactly as it was. New prompts in the file are added either way.
After every import the tool tells you how many were added, replaced, kept as yours, and already up to date.
Keeping it useful
- Agree the tag list early and write it down next to the master file. Left to itself, tagging produces
quiz,quizzesandassessmentfor the same thing within a week. - Prune monthly. The AI Assistant changes, and a prompt that worked in spring can produce junk by autumn; anything older than six months gets re-run or dropped.
- Forty good prompts beat two hundred near-duplicates. Edit the existing one.
- Put the failure in the notes. "Breaks the longest-answer rule about one time in five, check it" saves the next person the surprise.
- Tag project-specific prompts with the project name so they can be filtered out of the general list, or keep a separate JSON file per client.
Notes per target
Rise 360 code block
State the constraints every time: single file, nothing external, keyboard operable, prefers-reduced-motion respected, and the document must post its own height to the parent because the block sits in an iframe. Ask for "only the HTML" or you get an essay around it. Check the result at phone width before pasting it in; column layouts and card heights are the usual casualties.
Storyline 360 JavaScript
Name the Storyline variables in the prompt exactly as they appear in Storyline, including case, and say whether each is Text, Number or True/False. Ask for plain ES5 and no libraries. Test in a published output, not in Storyline preview, because the player API behaves differently there.
AI Assistant: images
Consistency comes from reusing identical wording. Keep everything after the subject the same across a course (palette, style, background, "no text, no logos") and change only the subject. Say the palette in words. Generate a few and pick; do not expect the first one to land.
AI Assistant: text
Paste source material into a placeholder rather than describing it. Give the output format explicitly (numbered questions, lettered options, correct letter, feedback line) and the rules the content must obey. Then check the rules yourself; instructions like "the correct answer must not be the longest" are followed most of the time, not every time.
Where the data lives
In your browser's local storage, tied to the address you opened the tool from. Two consequences:
- Opening the tool from a different address (the hosted link versus a downloaded file, or two different browsers) gives you two separate libraries. Pick one and stay with it, or move prompts between them with export and import.
- Clearing site data, or an IT policy that wipes browser storage, removes the library. Export after any session where you added something and treat the JSON file as the real copy.
Nothing you type leaves your machine. There is no telemetry and no network request beyond the fonts.