Last updated: 16 October 2026
Prompting Wizard is published by Entuvo. It is a set of files you install into an AI assistant you already use. Entuvo does not operate a Prompting Wizard account system or website backend. The course itself sends Entuvo nothing.
What we collect
Entuvo does not collect your lesson answers, prompts, scores, Learning Card, or PROGRESS.md contents.
Entuvo handles data about you in only two places, both described below:
- visit counts for the website, through Cloudflare Web Analytics;
- anything you choose to post in a public GitHub issue.
Your learning data stays with you and with the AI provider whose app or tool you use. That provider decides how long it keeps your chats and uploads. Entuvo cannot see or delete them.
The standalone course uses a local or attached PROGRESS.md. Optional local coaching stores learner data in an owner-only SQLite profile directory you choose. Entuvo receives neither file.
The website
The website (entuvo.github.io/prompting-wizard) counts visits with Cloudflare Web Analytics.
- What is recorded: each page view, with the page, the referring site, your country, device type, browser, operating system, and page load timings. It sets no cookies and uses no local storage. Cloudflare says it does not fingerprint visitors by IP address or user agent, and does not log query strings.
- Why: so Entuvo can see which pages people read and which links bring them to the course.
- Who receives it: Cloudflare, which runs the service under the Cloudflare privacy policy. Entuvo sees only aggregate counts in Cloudflare's dashboard, never an individual visit.
- How long it is kept: Cloudflare sets this, not Entuvo. Its Web Analytics FAQ says it keeps unsampled data for 7 days, then keeps about 10% of it for long-term storage, and the dashboard shows the previous six months.
- Your controls: the counter is a script from
static.cloudflareinsights.com. Block it, for example with a content blocker, and your visit is not counted. The site works without it.
The site is hosted by GitHub Pages, which keeps its own server logs under GitHub's policies. Videos load from YouTube only when you press play.
Optional local coaching
Only explicitly enabled workspaces are captured. Prompts and results are bounded to 16 KiB each, with truncation and incomplete-context labels. At most 1,000 completed pending examples are stored; a full queue skips new ones with a health warning. Pausing stops collection. Exclusion covers the selected workspace and descendants.
Raw examples and unreviewed findings expire after thirty days. Expired evidence is unavailable on runtime reads; physical cleanup happens on the next runtime invocation. A learner-confirmed lesson removes its source raw examples and findings earlier, retaining only the approved concise record. Reviewed records and imported course history remain until deleted. Original Day 0 baselines are preserved.
Both local hosts can use the same directory. When a lesson reads examples from another host, those excerpts enter the current provider's conversation and are subject to its policies. After-result coaching adds model inference and quota use; lesson-only collection does not. There is no Prompting Wizard cloud sync, account or telemetry service.
Use the runtime to inspect, exclude, pause or delete managed learning data. Deletion includes associated pending findings and drafts. Host transcripts, terminal/tool histories, process arguments, explicit exports and external backups are separate; the project cannot erase them or promise forensic erasure. Use stdin rather than command arguments for sensitive record text. See controls and setup.
Learning from past work history
A review of your past work runs only when you ask, one time. How your data is handled depends on where you run it.
Each provider's history is reviewed only by that provider's hosts: Claude Code and the Claude app review only Claude exports, and Codex and ChatGPT review only ChatGPT exports. Another provider's export is declined without review.
Past work history in Claude Code and Codex
pw history reads one of the following:
- your Claude Code or Codex sessions, through that host's documented SDK, skipping workspaces excluded from local coaching;
- a ChatGPT or Claude data export that you downloaded and chose to supply.
The prompting-wizard-openai.zip download reads exports only.
It never parses private transcript files. An export archive is read in memory and never extracted. You confirm twice: once before anything is read, and again before scoring. The runtime's shape check runs on your machine and stops another provider's export before anything reaches the model.
Scoring sends a sample of your prompts and the assistant's replies (300 by default) through the AI provider you are already using, under its terms, and uses your quota. Secrets that look like API keys, tokens, long hex strings or email addresses are masked before anything is written or sent. Pattern masking cannot catch everything.
Results are written under history/ in a local data directory you choose, with owner-only permissions. They include verbatim prompts, the casebook pages and an aggregate baseline. Runs are deleted 30 days after they are written, on the next runtime invocation, or sooner with delete.
Past work history in the ChatGPT and Claude apps
You upload an export to a chat and ask the coach to review it. There is no local runtime here, so the two confirmations and the secret masking above do not apply. The uploaded file goes to that app's provider as it is, with nothing masked, before the coach can check it or decline it. Upload only that app's own export, and remove anything you do not want the provider to see before you upload.
The coach reviews the file in that chat only. Prompting Wizard stores nothing. Findings appear in the reply, and the uploaded file stays in that chat under the provider's retention.
After any history review
Nothing is added to PROGRESS.md, a Learning Card, CLAUDE.md, AGENTS.md or any settings file unless you accept it. A Learning Card entry contains a paraphrase, never an excerpt. Entuvo receives nothing. Host transcripts and provider exports are governed by their own retention, and this project cannot delete them.
What the course may fetch
A locally loaded copy may read the public VERSION.md file from this GitHub repository to see whether a newer version exists. Automatic checks happen at most once every seven days. An explicit "check for updates" request, or a missing or unreadable last-check date, can fetch again sooner. That request goes to GitHub, not to an Entuvo server.
The prompting-wizard-openai.zip download has no update check and makes no such fetch; it updates only through the app's plugin manager.
Marketplace-managed and skills-CLI installs still apply updates through those hosts. They may also make the same read-only VERSION.md fetch when a check is due.
GitHub, Claude, OpenAI, the skills CLI, and any other host you use may collect their own logs or install telemetry under their policies. Those are not Entuvo services.
Practice prompts
Lesson prompts run in your AI assistant, with that assistant's file and network access. Outputs stay in that session unless you save them.
Support through GitHub Issues
Support and questions about this policy go through GitHub Issues. There is no email queue.
- What is shared: whatever you write in an issue or comment. It is public: anyone can read it, and it shows your GitHub username and profile.
- Why: Entuvo reads issues to answer you and to fix the project.
- Who receives it: GitHub, which hosts issues under the GitHub General Privacy Statement, and anyone who reads the issue.
- How long it is kept: an issue stays until it is deleted.
- Your controls: you can edit your issue and edit or delete your own comments. GitHub keeps an edit history that anyone who can read the issue can see. Only the repository's admins can delete an issue, so ask in the issue and Entuvo can delete it.
Use synthetic examples. Never post a Learning Card, PROGRESS.md, prompts, transcripts or exports.
Portable Learning Cards
Everyday learning contains no Wizard cloud service or background native-chat capture. A card contains only accepted learning evidence and preferences; avoid secrets and raw work excerpts. Attaching it to ChatGPT or Claude sends it to that provider under your account settings and its terms. Exporting it does not update or erase older uploads, chat history, backups or provider memory. Corrections/retractions affect the card you retain; delete other copies separately. Existing local coaching capture, retention and opt-in remain unchanged.
The native teaching plugins bundle instructions, rubric references and an empty card. They contain no MCP service, lifecycle hooks, local database or telemetry. They do not read other chats. Selected work and attached cards are processed by the host provider. Installing or removing a plugin does not save or delete your retained cards or prior provider uploads.