Impeccable explained: design guidance inside an AI coding workflow
A practical Impeccable review exercise for one page
Produce an evidence-backed design change that another reviewer can accept or reject.
What you will learn
- Write the brief first
- Make one bounded change
- Close the review record
Before you start
- One screen and its user task
- Authority to inspect an agent skill and project hook
Produce an evidence-backed design change that another reviewer can accept or reject.
Key takeaways
- A useful brief names the user task.
- One accepted change needs before-and-after evidence.
- Unresolved user questions stay visible.
Write the brief first
Choose one page and describe the user task, current friction and visual constraints in PRODUCT.md. If evidence is missing, write the uncertainty rather than manufacturing a persona. Capture the current page at desktop and narrow viewport.
State whether the exercise is composition-first or code-first. A page with a stable component system may need a focused polish pass; a new surface may need a composition study before implementation.
Make one bounded change
Run a relevant command or a deterministic `detect` pass, then select one finding. Save the precise diff and a same-viewport screenshot after the change. Avoid accepting a whole bundle of unrelated edits because one rule sounded persuasive.
If the change affects a component reused elsewhere, inspect at least one other consumer. A local visual improvement can create a layout regression in another route.
Close the review record
Record what the detector reported, why the reviewer accepted or rejected the change, and what the application tests found. Note anything that still requires a user test.
This exercise is proposed by EasyAI, not an automatic Impeccable workflow. The series did not run the exercise on a production site; its value comes from the evidence a team can inspect.
Decision guide
| Criterion | Option A | Option B |
|---|---|---|
| Best when | You need predictable behavior and easy auditing | You need adaptive optimization and have reliable telemetry |
| Main risk | May leave performance on the table | Can become difficult to explain or debug |
Implementation steps
- 1
Describe one page and capture its baseline.
- 2
Apply one bounded change and preserve the diff.
- 3
Record reviewer decision and regression checks.
Copy-ready example
ui_review:
page: choose-one
baseline_viewports: [desktop, narrow]
detector_finding: record
accepted_diff: review
user_test: pendingFrequently asked questions
Does the tool create this review record automatically?
No. This is a suggested team exercise built around documented commands.
Can I skip responsive review?
A desktop-only screenshot cannot establish behavior on a narrow viewport.
Sources
- Impeccable / README.mdSource checked 2026-09-29
- Impeccable / docs/CLI-CONTRACT.mdSource checked 2026-09-29