ui-percept-rapid-feedback（ui 感知快速反馈）

UI Percept Rapid Feedback

Use this skill when you need quick iteration on Quick Maths screen quality, especially spacing, alignment, sizing consistency, truncation risk, and palette cohesion.

This skill avoids subjective pass/fail judgment and instead produces concrete, measurable feedback that an agent can use to patch SwiftUI code and rerun.

When To Use

• After UI changes in quick-math-swift
• When reviewing one or more captured screens (home, game, reward)
• When you want a higher aesthetic bar than default ui-percept rules
• When you need output saved to files for repeatable agent loops

Prerequisites

• quick-math-swift repo is present and buildable
• ui-percept repo exists at /Users/vabole/repos/ui-percept or passed via --ui-percept-dir
• iOS simulator + xcodebuild available for capture

Workflow

1. Run the helper from quick-math root:

bash .apple-skills/skills/ui-percept-rapid-feedback/scripts/run-feedback.sh --mode improve --screens game

1. Read:

• .ui-captures/ui-percept-skill/<run-name>/feedback-overview.json first
• .ui-captures/ui-percept-skill/<run-name>/codex-jury-prompt.md for Codex vision review mode
• .ui-captures/ui-percept-skill/<run-name>/codex-decision-log.md for narrated decision tracking
• .ui-captures/ui-percept-skill/<run-name>/feedback-report.md second
• .ui-captures/ui-percept-skill/<run-name>/feedback-report.json only when drilling down

1. Prioritize fixes in this order:

• blocker and major lint issues
• repeated rule clusters in report summary
• top suggestions from signals.json

1. Patch SwiftUI code, then rerun the same command.

Commands

Run selected screens:

bash .apple-skills/skills/ui-percept-rapid-feedback/scripts/run-feedback.sh --mode improve --screens home,game,reward

Switch to build-oriented prompt mode:

bash .apple-skills/skills/ui-percept-rapid-feedback/scripts/run-feedback.sh --mode build --screens game

Archive before/after screenshots with commit-hash labels:

# before changes (uses current HEAD hash by default)
bash .apple-skills/skills/ui-percept-rapid-feedback/scripts/run-feedback.sh \
  --mode improve --screens game --archive-stage before --archive-user "$USER"

# after changes (pass the new commit hash explicitly)
bash .apple-skills/skills/ui-percept-rapid-feedback/scripts/run-feedback.sh \
  --mode improve --screens game --archive-stage after --archive-user "$USER" --git-hash <new_commit_hash>

Low-friction helper (auto hash + optional history commit):

# before phase
bash .apple-skills/skills/ui-percept-rapid-feedback/scripts/archive-round.sh --phase before --screens game

# after phase
bash .apple-skills/skills/ui-percept-rapid-feedback/scripts/archive-round.sh --phase after --screens game

Skip automatic history commit if you only want artifacts:

bash .apple-skills/skills/ui-percept-rapid-feedback/scripts/archive-round.sh \
  --phase after --screens game --no-history-commit

Use existing captures without rerunning XCUITest:

bash .apple-skills/skills/ui-percept-rapid-feedback/scripts/run-feedback.sh --screens game --skip-capture

Set custom ui-percept path:

bash .apple-skills/skills/ui-percept-rapid-feedback/scripts/run-feedback.sh \
  --screens game \
  --ui-percept-dir /path/to/ui-percept

Analyze a specific archived artifact directly (no new capture):

bash .apple-skills/skills/ui-percept-rapid-feedback/scripts/run-feedback.sh \
  --tree .ui-captures/game/ui_tree.json \
  --screenshot .ui-captures/ui-percept-v2/game-fast/full.png \
  --screen-name game-fast \
  --run-name game-fast-review

Notes

• This skill uses references/high-bar-rules.yaml to raise visual standards.
• Keep feedback evidence-based: element IDs, measured values, and repeatable signals.
• Context hygiene rule: load feedback-overview.json first and only open per-screen drilldown files for the issue you are fixing.
• For opinionated UX/aesthetic review, apply references/jury-review-mode.md.
• For round-by-round history and commit discipline, apply references/round-commit-workflow.md.