tcr-apply

Run the skill from the repo root of your working app. It detects the UI pattern from your code and returns three sequential prompts — one each for Transparency, Control, and Recovery — that you paste into Claude Code one at a time, reviewing and committing after each.

Seven patterns are supported: chat with context panel, upload-to-dashboard, query-to-structured-result, wizard with inline audit, draft-approve-send, queue with approval, and real-time streaming. Detection uses code signals like st.chat_message, file uploaders, SQL runners, and EventSource usage.

No. Every generated prompt is explicitly additive — it adds new components and features without refactoring, restructuring, or moving existing files. Anti-patterns like bundling all three prompts into one or prescribing specific component libraries are forbidden.

If code signals are weak or ambiguous, the skill lists the seven pattern names and asks which one closest matches your app instead of guessing silently. A wrong pattern guess would produce wrong prompts, so it confirms first.

Yes. The skill reads your app's existing UI strings (button labels, placeholders, JSX copy) and mirrors that language in the generated prompts. If a GUIDELINE.md file exists from a prior ui-pattern run, it uses that to pick domain-specific copy and trust-signal types.

The T-C-R features are designed to compound sequentially. Pasting one prompt, reviewing the diff, testing in the browser, and committing before the next keeps each change reviewable and preserves the pedagogical rhythm of the workshop.