animate (emilkowalski/skills)

评测

在 Codex CLI（SOTA 档）上跑 3 个任务、对照、一句话提示、装 skill 三组，共 18 次运行，由规则脚本打分（没有 LLM 评委），断网沙箱、每次运行独立临时目录。具体做法见下方英文评测原文。

三组对比

Codex CLI 实测，18 次运行，规则脚本打分，没有 LLM 评委。总分 22.0 / 30，结论 视情况而定。

下面是评测原文，英文。

Verdict: Situational (22.0 / 30). Source: emilkowalski/skills, skills/animate, pinned at commit a81dbca. Install: npx skills@latest add emilkowalski/skills --skill animate. Licence: MIT (repo LICENSE).

What it is

The animate skill (emilkowalski/skills): motion rules for building UI animations: animate only with purpose, custom ease-out cubic-bezier (never ease-in or the default), 150-300 ms, transform and opacity only, press feedback with scale(0.97), never start from scale(0), stagger entrances, interruptible transitions, prefers-reduced-motion, hover gated with (hover: hover).

How we tested

Codex CLI 0.166 (gpt-6-astra, low reasoning) on a dedicated Windows test PC, a fresh temporary project and temporary HOME/TMPDIR/caches per run, skill installed (treatment) or absent (control) or absent with the skill's core requirement squeezed into one sentence appended to the prompt (one-liner group), 2 runs per group per task, network off, /tmp writes blocked. The prompts never mention the skill (except the wording noted below).

Three build tasks, each one self-contained HTML file with inline CSS and JS: a dropdown menu, a modal with a dimmed backdrop, and stacking toasts, all with open/close or enter/exit animations. Prompts do not mention motion rules. The one-liner lists the rules in one sentence.

1. Animated dropdown menu — Build menu.html: a button that opens a dropdown menu with 5 items, with open and close animations. Write one self-contained HTML file (inline CSS and JS, no libraries).
2. Animated modal dialog — Build modal.html: a button that opens a modal dialog with a dimmed backdrop, with open and close animations. Write one self-contained HTML file (inline CSS and JS, no libraries).
3. Toast notifications — Build toasts.html: three buttons that each show a toast notification which stacks and dismisses itself, with enter and exit animations. Write one self-contained HTML file (inline CSS and JS, no libraries).

Scoring is programmatic (hidden tests / planted-issue recall / constraint checks / render+OCR where relevant), 0–5 per run. No LLM judge decided any score.

Results

The one-liner given to the third group: "Animate like a design engineer: only animate what has a purpose, use a custom ease-out cubic-bezier (never ease-in or the default), 150-300 ms, only transform and opacity, press feedback with scale(0.97), never start from scale(0), stagger entrances, interruptible transitions, honour prefers-reduced-motion and gate hover with (hover: hover)."

Mean score: 3.71 without vs 4.36 with the skill (delta +0.64). Tokens per run: 78k vs 132k (1.68x); time 82s vs 81s. One-liner group: 5.0 (the skill's gain over the one-liner: -0.64; one-liner's own gain over no skill: +1.29); one-liner tokens 105k per run, so the skill costs 1.25x the one-liner.

What happened

• Control 3.71, skill 4.36, one-liner 5.0 (n=6 each). A rule-based motion check counts how many of 10 rules the page follows. Control reached 6-8 of 10, the skill 8-9, and the one-liner satisfied every check in all 6 runs.
• The rules control most often missed: press feedback (an :active scale) in 3 of 6 runs, plus staggered entrances and non-default easing in some runs. The skill still missed press feedback in 2 runs and staggered entrance or easing in several; on the toast task its score was 4.14 and 4.57 against 4.14 for control.
• Tokens: skill 132k, one-liner 105k, control 78k per run (1.7x control, 1.3x one-liner). The skill was read in 6/6 treatment runs; it uses more commands (9.3 per run against 4.7).
• Treatment runs add about 3 new files (the skill copy in the working folder) per run.

Dimension scores

Limits of this test

• Rule-based checks on the code text of one HTML file; the animations were not rendered or timed in a browser, so smoothness and visual quality are not judged.
• The one-liner is a close paraphrase of the skill's own checklist, and the checks mirror the same rule list.
• n=2 per cell.
• n = 2 per group per task: differences under about 1 point are noise.
• Rule-based scoring measures what the checks can see; it cannot judge taste or whether a result is the best possible.

Security

See security.md. Scan: 0 CRITICAL / 0 HIGH / 0 MEDIUM. Verdict: safe to run (no-network sandbox). A clean scan does not prove a skill safe.

Bottom line

The skill improves the motion rules followed over no guidance, but the same rules as one sentence achieve a perfect score for fewer tokens, so the benefit is over a bare prompt only.