Skip to content

Feature field guide

The feature field guide shows seven example scenes with checkpoints and expected results. Every pull request renders GIF/PNG previews using the CLI built from that pull request.

  • From intent to input: environment values, typing, clipboard editing and encoded keys.
  • Text with character: emphasis, truecolor, Unicode and cell graphics.
  • Room to breathe, dark and light: matched content with themes and window framing.
  • Present the action: captions and keyboard chips, with clean and annotated checkpoints.
  • Reveal the result: hidden setup remains visible to checkpoints but absent from animation.
  • Keep the history: scrollback and alternate-screen restoration.

In a pull request’s CI → Package and smoke (ubuntu-24.04) job summary, choose Betamax field guide. The standalone HTML includes animations, comparisons, descriptions, expectations and tapes. The feature-checkpoints artifact contains the PNG/GIF/state JSON files. GitHub sign-in is required; artifacts expire with retention.

To generate a guide locally from the repository root:

Terminal window
mise run feature-gallery

Open target/features/feature-gallery.html in a browser. Python 3, ffmpeg and the repository’s mise build tools are required. The runner supplies the fixture’s absolute path to the PTY environment. Generated files stay outside version control.

The suite checks exact input and state, image regions, paired checkpoints and every hidden-setup animation frame. It avoids whole-image comparisons across different platform fonts. A separate three-second playback tape checks GIF/PNG/MP4/WebM, while Linux/macOS renderer tests keep their specialized glyph and cell diagnostics.

There is a known underline/strikethrough rendering gap: state spans are correct, but those decorations are absent from the image. The gallery labels this explicitly. Passing the current checks does not mean every style or every parameter value is covered.

Read the complete coverage map and comparison policy before extending the suite. The feature tapes and scene catalog keep names and expected results close to the rendered evidence.