Skip to content

Privacy and honesty ​

What never goes on screen ​

  • secrets: .env values, API keys, tokens, passwords
  • internal hostnames, IP addresses, localhost URLs and stack traces
  • real customer data and the names of your other clients
  • file names, function names and commit hashes, unless you ask for audience=technical

Screenshots use demo data. Emails, phone numbers, card numbers, tokens and password fields are covered automatically, and you can mask more by selector.

Before anything is handed over, the QA script scans the captions, slide copy, every diagram label and the message for API keys, tokens, JWTs, private keys, credential assignments, real email addresses and internal hosts. Diagram sources are scanned line by line, comments included, so a label in unusual Mermaid syntax is still checked. Claude then looks at every extracted frame. A secret drawn inside an image or canvas in your own app is caught by that visual check alone, so keep demo data in the app you capture.

What is never invented ​

The skill describes only what the diff shows. It does not make up features, numbers, dates or screens. Visuals that are not real screenshots stay clearly illustrative, and limitations and required actions are stated plainly, including "nothing to do on your side".

Your project stays untouched ​

The skill reads your project and writes only into the output folder and its own tool cache. It never edits, commits, stashes or pushes your code and never installs dependencies into your project. A "before" screenshot uses a temporary git worktree that is removed afterwards.

Network ​

Rendering is local. The first run downloads the tools once. Telemetry in HyperFrames is turned off. The skills CLI you install with sends anonymous install counts for public repositories; set DISABLE_TELEMETRY=1 to turn that off.