---
name: optimize-tui
description: Use this skill when an agent must design, implement, validate, or review a terminal user interface; capture reproducible terminal screenshots or animations; package PNG, GIF, or MP4 evidence; iterate with advanced terminal, design, accessibility, and reliability personas; or publish visual evidence to a pull request. Activate for TUI, terminal dashboard, animated CLI, alternate screen, resize/redraw jitter, terminal evidence, or visual PR review work.
license: MIT
compatibility: Requires this repository, Git, GitHub CLI authentication for PR publishing, pnpm, Python with Pillow for evidence capture, a Chromium browser, and ffmpeg only when MP4 output is needed.
metadata:
  author: MSFT-TKENDRICK
  version: '0.1.0'
---

# Optimize TUI

Use executed synthetic scenarios through the production orchestrator and renderer as the source of
truth. Direct renderer fixtures are not acceptance evidence. A TUI change is not complete until
behavior, executable trace, visual evidence, and pull-request review material agree.

## Route the task

Read only the reference needed for the current phase:

- Read [iterative workflow](references/workflow.md) for every new or resumed TUI change.
- Read [evidence and media](references/evidence-and-media.md) before capturing, compressing, or
  selecting PNG, GIF, or MP4 assets.
- Read [quality and convergence](references/quality-and-convergence.md) before accepting a visual
  iteration or completing adversarial review.
- Read [pull-request evidence](references/pull-request-evidence.md) before committing assets,
  uploading media, or updating a pull request.

## Required behavior

1. Capture a source-bound baseline before editing.
2. Update production behavior, focused tests, Gherkin scenarios, and persona expectations together.
3. Regenerate deterministic synthetic evidence and its execution manifest after each material visual
   iteration; fail closed if any asset lacks an executed scenario event.
4. Inspect every state and animation; continue until persona, designer, accessibility, and adversarial
   review have no unresolved material finding.
5. Run focused gates, `npm run test:bdd`, `npm run optimize:ux -- cycle` when modeled journeys
   changed, and `npm run check`.
6. Commit the latest reviewable evidence and update the pull request so reviewers need not run the CLI.

## Non-negotiable safety

- Never put credentials, tenant data, migration reports, or non-synthetic traces in visual evidence.
- Never hide jitter, clipping, lifecycle leaks, or accessibility regressions by weakening tests.
- Never send binary media or base64 through the conversation; upload compressed files directly.
- Keep every CAPI-bound asset below 5 MiB, with a target maximum of 4.5 MiB.