ux-critique

スター0

フォーク0

更新日2026年7月31日 08:07

UX design critique with staff-level evaluation. Use when reviewing UI/UX, critiquing designs, or before shipping a UI feature.

インストール

Codex または Claude でインストールこの Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。

ソース

ugudlado

ugudlado/orchestrator-pack

GitHub リポジトリを開く Creator のリポジトリを見る

ダウンロード

ファイルエクスプローラー

6 ファイル

SKILL.md

readonly

name	ux-critique
description	UX design critique with staff-level evaluation. Use when reviewing UI/UX, critiquing designs, or before shipping a UI feature.
user-invocable	true
extends	git+git@github.com:ugudlado/prompt-packs.git@302b87dcc7c8b6a83d249194f3e47e98d3214794#ux-reviewer

Run UX Critique

Intent: Run UX critique on UI changes and iterate until the target score is met.

Inputs

None named. (Reads modified files in the phase.)

Outputs

critique_score
critique_skipped
critique_retries

Instructions

Quality thresholds are step-owned (vendor this pack to change them):
- target_score = 8
- max_retries = 3
Check if any files modified in this phase touch UI: Match: _.html, _.css, _.scss, _.tsx, _.jsx, _.svelte, _.vue, _.astro, or files in components/, pages/, views/, layouts/, templates/.

If NO UI files modified → skip. Log: "[critique] No UI changes — skipping"
Perform UX critique directly on:
- Target files (list of modified UI files)
- target users, as stated in the repo's own docs (README / CLAUDE.md), if any
- the step-owned scoring thresholds (critical_cap 5, important_cap 7, green_base 9)
- ux-prototype.html reference if it exists in change dir
Read SCORE and STATUS from your critique output.
If score >= target_score: PASS. Record critique_score in state.yaml.
If score < target_score: a. Parse PRIORITY_ISSUES from agent output into fix tasks. b. Apply fixes scoped to critique findings only. c. Run the repo's verify commands (as discovered from CLAUDE.md / AGENTS.md / README / manifest conventions) to confirm nothing broke. d. Increment retry counter in state.yaml. e. Re-run critique on updated files. f. If retries >= max_retries: STOP and escalate to user.

Commit UX improvements:

style(<change-id>): UX critique improvements (score: N/10)

Rules (constraints on how)

Only runs when the phase includes UI-facing changes.
Target score is 8 (step-owned).
Retry with fixes until target score is met or max_retry_rounds exhausted.
Perform critique using this skill procedure.

Verify

If UI files modified: critique_score >= target_score (8)
If no UI files: step skipped (logged)
All discovered verify commands pass after fixes

Return COMPLETION

After verify passes (or on skip), return:

COMPLETION:
  status: completed
  outputs:
    critique_score: <N or null if skipped>
    critique_skipped: <true if no UI files, false otherwise>
    critique_retries: <N>

このリポジトリの他の Skills

同じリポジトリ

explore

ugudlado/orchestrator-pack

Explore the codebase and write a discovery brief. Use when discovering scope, surveying a codebase, or starting a feature.

2026-07-310

approve-qa

ugudlado/orchestrator-pack

QA passed — merge branch to main, move ticket to Done, delete branch. Use after QA sign-off on a completed feature.

2026-07-310

autopilot

ugudlado/orchestrator-pack

Run one fully-autonomous development iteration. Picks a ticket (from --focus or backlog via ideator), then runs the `orchestrator autopilot <slug>` workflow subcommand. This skill should be used when the user says 'autopilot', 'autonomous', 'self-improve'.

2026-07-310

commit-group

ugudlado/orchestrator-pack

Create commits in logical groups. This skill should be used when the user says 'commit-group', 'group commits', 'organize commits', or when there are multiple unstaged changes that should be organized into atomic commits.

2026-07-310

complete-feature

ugudlado/orchestrator-pack

Complete feature — verify, signoff, archive. Runs only the complete phase of the orchestrate workflow. This skill should be used when the user says 'complete feature', 'finish feature', 'merge to main'.

2026-07-310

context-hub

ugudlado/orchestrator-pack

Fetch curated, agent-optimized library documentation via Context Hub (chub). Use when implementing features that use external libraries, when Context7 returns noisy/incomplete docs, when you need language-specific or version-specific API docs, or when the user says "chub", "context hub", "get docs for X". Prefer this over Context7 for libraries in chub's registry — curated docs have less noise and better code examples.

2026-07-310

Run UX Critique

Intent: Run UX critique on UI changes and iterate until the target score is met.

Inputs

None named. (Reads modified files in the phase.)

Outputs

critique_score

critique_skipped

critique_retries

Instructions

Quality thresholds are step-owned (vendor this pack to change them):

target_score = 8
max_retries = 3

Check if any files modified in this phase touch UI: Match: _.html, _.css, _.scss, _.tsx, _.jsx, _.svelte, _.vue, _.astro, or files in components/, pages/, views/, layouts/, templates/.

If NO UI files modified → skip. Log: "[critique] No UI changes — skipping"

Perform UX critique directly on:

Target files (list of modified UI files)
target users, as stated in the repo's own docs (README / CLAUDE.md), if any
the step-owned scoring thresholds (critical_cap 5, important_cap 7, green_base 9)
ux-prototype.html reference if it exists in change dir

Read SCORE and STATUS from your critique output.

If score >= target_score: PASS. Record critique_score in state.yaml.

If score < target_score: a. Parse PRIORITY_ISSUES from agent output into fix tasks. b. Apply fixes scoped to critique findings only. c. Run the repo's verify commands (as discovered from CLAUDE.md / AGENTS.md / README / manifest conventions) to confirm nothing broke. d. Increment retry counter in state.yaml. e. Re-run critique on updated files. f. If retries >= max_retries: STOP and escalate to user.

Commit UX improvements:

style(<change-id>): UX critique improvements (score: N/10)

Rules (constraints on how)

Only runs when the phase includes UI-facing changes.

Target score is 8 (step-owned).

Retry with fixes until target score is met or max_retry_rounds exhausted.

Perform critique using this skill procedure.

Verify

If UI files modified: critique_score >= target_score (8)

If no UI files: step skipped (logged)

All discovered verify commands pass after fixes

Return COMPLETION

After verify passes (or on skip), return:

COMPLETION: status: completed outputs: critique_score: <N or null if skipped> critique_skipped: <true if no UI files, false otherwise> critique_retries: <N>