Review
Read-only judgment of existing UI. This spine routes; the judgment lives in the
reference files — load only what the mode needs.
Contract (all modes)
- Read-only. Never edit product code, even for a one-character fix. Findings name
the change; applying it is the pe-build skill's job — hand off by name.
- Engine rules govern every mode. Read
references/engine.md first, always: its
evidence bar, escalation triggers, cheaper-fix ladder, consolidation, and the
verdict format in references/review-format.md.
- Severity is HIGH / MEDIUM / LOW plus the engine's escalation triggers.
Accessibility findings are never discarded.
Modes
| Mode |
When |
Load |
| change |
A diff, branch, PR, or "review my changes" is named. Always wins over screen when both could apply. |
change-review.md, scope-resolution.md, removed-signals.md |
| screen |
A screen, page, site, or screenshot; "review this UI." |
screen-review.md, experience-rubric.md, technical-rubric.md |
| improve |
"Audit this and give me a roadmap", "improve the app" — whole-surface improvement, delivered as self-contained plans another agent executes. Explicit invocation only. |
audit-method.md, plan-template.md |
| guidelines |
"Check against best practices / the guidelines"; terse file:line lint of UI code. |
web-interface-guidelines.md |
| stress |
"Does this component survive?" — render one component in every hostile state. Explicit invocation only. |
stress.md, stress-scenarios.md |
| motion |
"Review the animations / motion." For a whole-codebase motion audit with plans, add motion-audit.md + motion-audit-signals.md + plan-template.md. |
motion-review.md, motion-standards.md |
| opportunities |
"What could be animated here?" — read-only; proposes motion with exact values, never implements. Explicit invocation only. |
motion-opportunities.md |
| a11y |
"Is this accessible?" — audit, not implementation. |
a11y-audit.md (+ engine triggers) |
| fidelity |
"Did we stay true to the mock / design?" — the implementation against the approved artifact in .product/approved/; every deviation is a finding with severity. Implied whenever a reviewed surface has an approved record. |
fidelity.md |
Disambiguation: verb beats noun — "review the animation" is motion mode here;
"fix the animation" is not this skill at all (pe-build). A request that mixes modes runs
the engine once and consolidates, never two parallel reviews.
Name mapping
Some references route to skills by name. Those names resolve as follows — apply the
mapping silently:
| Text says |
Use |
| better-accessibility |
a11y-audit.md for ordering; the engine's own triggers for severity; the pe-build skill's references/a11y/ for depth |
| better-ui, better-layout, better-typography, better-colors, better-writing |
the pe-build skill's references/craft/{ui,layout,typography,colors,writing}/ |
| better-interface |
this skill's engine mode itself |
House rules
- Autocomplete: on for purposeful fields;
off only for non-auth-sensitive cases.
- P0–P3 appears only inside screen mode's files; report severity as HIGH/MED/LOW
(P0+P1 → HIGH, P2 → MEDIUM, P3 → LOW).
Handoffs
Fixes → pe-build (findings include exact file:line and the named change).
Proof that behavior works end to end, with recordings → pe-verify; this skill judges
quality, that one runs the checks.
Documenting the system a review revealed → pe-design, understand mode.
Never resolve a finding by editing anything yourself.
1---2name: pe-review3description: Read-only review of existing UI with verdicts. Use to critique a screen, site, or screenshot; review a diff, branch, or PR for interface quality; produce a prioritized audit with implementation plans for another agent; check code against the Web Interface Guidelines; stress-test a component in hostile states; review animation and motion craft; audit accessibility; or check an implementation against the approved design record. Triggers on review, critique, audit, UX review, design review, interface review, "review my PR", "check accessibility", "will this survive", "review the animations", "did we match the mock". Produces findings and verdicts only — never edits code.4license: Apache-2.05---67# Review89Read-only judgment of existing UI. This spine routes; the judgment lives in the10reference files — load only what the mode needs.1112## Contract (all modes)1314- **Read-only.** Never edit product code, even for a one-character fix. Findings name15 the change; applying it is the pe-build skill's job — hand off by name.16- **Engine rules govern every mode.** Read `references/engine.md` first, always: its17 evidence bar, escalation triggers, cheaper-fix ladder, consolidation, and the18 verdict format in `references/review-format.md`.19- **Severity** is HIGH / MEDIUM / LOW plus the engine's escalation triggers.20 Accessibility findings are never discarded.2122## Modes2324| Mode | When | Load |25| --- | --- | --- |26| **change** | A diff, branch, PR, or "review my changes" is named. Always wins over screen when both could apply. | `change-review.md`, `scope-resolution.md`, `removed-signals.md` |27| **screen** | A screen, page, site, or screenshot; "review this UI." | `screen-review.md`, `experience-rubric.md`, `technical-rubric.md` |28| **improve** | "Audit this and give me a roadmap", "improve the app" — whole-surface improvement, delivered as self-contained plans another agent executes. Explicit invocation only. | `audit-method.md`, `plan-template.md` |29| **guidelines** | "Check against best practices / the guidelines"; terse file:line lint of UI code. | `web-interface-guidelines.md` |30| **stress** | "Does this component survive?" — render one component in every hostile state. Explicit invocation only. | `stress.md`, `stress-scenarios.md` |31| **motion** | "Review the animations / motion." For a whole-codebase motion audit with plans, add `motion-audit.md` + `motion-audit-signals.md` + `plan-template.md`. | `motion-review.md`, `motion-standards.md` |32| **opportunities** | "What could be animated here?" — read-only; proposes motion with exact values, never implements. Explicit invocation only. | `motion-opportunities.md` |33| **a11y** | "Is this accessible?" — audit, not implementation. | `a11y-audit.md` (+ engine triggers) |34| **fidelity** | "Did we stay true to the mock / design?" — the implementation against the approved artifact in `.product/approved/`; every deviation is a finding with severity. Implied whenever a reviewed surface has an approved record. | `fidelity.md` |3536Disambiguation: **verb beats noun** — "review the animation" is motion mode here;37"fix the animation" is not this skill at all (pe-build). A request that mixes modes runs38the engine once and consolidates, never two parallel reviews.3940## Name mapping4142Some references route to skills by name. Those names resolve as follows — apply the43mapping silently:4445| Text says | Use |46| --- | --- |47| better-accessibility | `a11y-audit.md` for ordering; the engine's own triggers for severity; the pe-build skill's `references/a11y/` for depth |48| better-ui, better-layout, better-typography, better-colors, better-writing | the pe-build skill's `references/craft/{ui,layout,typography,colors,writing}/` |49| better-interface | this skill's engine mode itself |5051## House rules5253- Autocomplete: on for purposeful fields; `off` only for non-auth-sensitive cases.54- P0–P3 appears only inside screen mode's files; report severity as HIGH/MED/LOW55 (P0+P1 → HIGH, P2 → MEDIUM, P3 → LOW).5657## Handoffs5859Fixes → **pe-build** (findings include exact file:line and the named change).60Proof that behavior works end to end, with recordings → **pe-verify**; this skill judges61quality, that one runs the checks.62Documenting the system a review revealed → **pe-design**, understand mode.63Never resolve a finding by editing anything yourself.
backnotprop