| name | uno-review |
| description | Diagnose-only design review with three scenarios: (1) stage-lens review of any artifact exiting prototyping — DS, product-intent (UNO), and accessibility lenses in parallel at stage-appropriate depth; (2) the handoff gate (DS / UNO / a11y) before a package publishes; (3) Design QA — Figma spec vs the QA build when a Roadmap card hits Ready for QA. Use when the user says "review this", "critique", "check against the design system", "poke holes", "run design QA", when an artifact exits uno-prototype, before uno-publish hands off, or when Dev Status flips to Ready for QA (RTT). Mandatory input: a one-line artifact manifest (fidelity / tools / PRD link). It never fixes what it finds — fixes route to uno-prototype (artifact) or uno-maintain (harness/docs).
|
| user-invocable | true |
| argument-hint | ["artifact + manifest (fidelity / tools / PRD link)"] |
| allowed-tools | Read, Grep, Glob, Bash, Task |
Quality Review
Agents it summons
reviewers/ds-lens · reviewers/uno-lens · reviewers/a11y-lens · reviewers/design-qa · reviewers/rubric-applier — defined in agents/ (see agents/README.md). Per the interaction contract, these are summoned by this skill, never by users.
Poke holes in a design with a stage-aware lens. Diagnose only — this skill never edits the artifact. The full procedure (intake, scenarios, lens depth, severity, verdict) is references/method.md — normative for both faces; this file adds only IDE execution.
When it fires
- Stage-lens review — an artifact exits any
uno-prototype path (including hand-crafted work), or the user asks for a review/critique.
- Handoff gate — a handoff package is about to publish via
uno-publish: after rails propagation, before sign-off.
- Design QA — auto-triggered when the Roadmap card hits
Dev Status: Ready for QA (RTT); also on request ("run design QA on X").
Suggest proactively when the user says "done", "ready to share", or is about to invoke uno-publish — once per work session, don't repeat if declined.
Workflow
- Collect the manifest — fidelity (low / mid / high / coded) · tools · PRD link. Missing → ask for it; no manifest, no review (method.md § Intake).
- Load
references/method.md and identify the scenario (method.md § Three scenarios).
- Dispatch the lenses:
- Stage-lens / handoff gate → summon reviewers/ds-lens + reviewers/uno-lens + reviewers/a11y-lens in parallel, each with the artifact + manifest. Each stays in-lane at the manifest's depth — no token nits on low-fi work.
- Design QA → summon reviewers/design-qa with the Roadmap card; it resolves RM-ID →
[spec] Figma file and walks the QA build against it with the Design QA checklist.
- Coded artifacts: run
bash skills/uno-review/scripts/run-review-checks.sh <dir> and hand the hits to ds-lens as evidence (patterns: references/catch-patterns.md).
- Merge findings — dedupe cross-lens overlaps, keep each finding's severity · lens · evidence · reference · re-entry point (method.md § Findings & severity).
- Verdict per method.md § Verdict & re-entry:
Issues? = Yes only at severity major+; minors travel as advisory; Design QA blockers hold Ready for Prod.
- Log the run — summon reviewers/rubric-applier with
docs/evals/rubrics/uno-review.md; it scores the dimensions and appends the eval-run entry. Golden scenarios this skill must pass: docs/evals/scenarios/uno-review.md.
Tier-2 loads
| Trigger | Load |
|---|
| always | references/method.md |
| coded artifact | references/catch-patterns.md (+ the script above) · sample report shape: examples/review-output-example.md |
| artifact touches Storybook stories/MDX | references/storybook.md · references/storybook-component-docs.md |
| posting verdicts / sign-offs to Slack | docs/conventions/slack.md |
Lens rule docs (DS agent-views, foundations/accessibility.md, figma-workspace.md) are loaded by the summoned agents, not here.
Quality bar
Scored by docs/evals/rubrics/uno-review.md: recall ≥80% and precision ≥80% on the golden defect set, 0 out-of-lens findings, diagnose-only hard gate. Golden scenarios: docs/evals/scenarios/uno-review.md.
Constraints
- Never edits the artifact, never offers to fix — 0 tolerance. Fixes are explicit, separate asks: artifact →
uno-prototype, harness/DS/docs → uno-maintain.
- One artifact per review; product-direction questions ("should this exist?") escalate to Bill.
- Skill-definition quality audits are not this skill — that path lives in
uno-maintain (skills/uno-maintain/references/skill-quality/).
Re-entry
Issues? = Yes → back to the fidelity decision in uno-prototype; each finding names its suggested re-entry point.- Clean handoff gate → sign-off, then
uno-publish proceeds.
- Design QA findings → annotated issues to dev before
Ready for Prod.
- Lesson worth keeping surfaced by the review →
uno-maintain knowledge capture.
