菜单
Human-in-the-loop web studio to tune AI-generated output by eye. Stand up a local interactive studio (sliders, pickers, drag handles) or an inline edit/highlight/comment annotation studio for prose &
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/ranbot-ai/awesome-skills --skill lookdev命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
基于 SOC 职业分类
正在显示 SKILL.md
| name | lookdev |
| description | Human-in-the-loop web studio to tune AI-generated output by eye. Stand up a local interactive studio (sliders, pickers, drag handles) or an inline edit/highlight/comment annotation studio for prose & |
| category | Document Processing |
| source | antigravity |
| tags | ["python","react","node","markdown","api","claude","ai","agent","workflow","design"] |
| url | https://github.com/sickn33/antigravity-awesome-skills/tree/main/skills/lookdev |
Use when the user says "lookdev", or asks to tune / dial in / iterate on the look of something, compare variations by feel, or review / edit / annotate a blog post, doc, copy, or media set. Use whenever "show me, I'll pick" beats asking the user to specify a number, and whenever you'd otherwise hand back a static grid or a wall of prose for review.
Source: connerkward/lookdev-studio-skill (MIT).
When the user says "lookdev" — or any of: tune, , , , , , — they mean . Not a static grid of N variations. Not a Q&A where they specify numbers. Not a wall of prose they're asked to read and reply to in chat. A real-time studio where they act on the artifact and the change is captured.
Two studio shapes — pick by what's being tuned:
Any visual decision the user picks by feel, not by spec. Expand this list as needed:
If the studio shows a list, grid, or scroll-long set of variations, controls must be visible from every scroll position. The user has to be able to drag a slider while looking at row 14, not scroll back to the top each time.
Two approaches, pick by layout:
position: sticky; top: 0) at the top of the scroll container. Keep the bar visually distinct — paper background + blur backdrop + bottom border — so it doesn't muddy the specimens scrolling behind it. Sticky pins relative to the nearest scrolling ancestor with a defined boundary; if you nest it inside a sized parent (a <header> with margin-bottom, a <div> with a fixed height), it stops sticking at that parent's bottom edge. Lift it to be a direct child of <body> (or the page-wrap) so stickiness spans the whole page.position: fixed) for hotkey-toggled controls — e.g. press d to reveal. The portfolio's .debug-ctl pattern is this: pinned top-left, transparent until summoned. Use when the controls shouldn't occupy permanent screen real estate (final viewers shouldn't see them; the author can summon on demand).Anti-pattern: a top-of-page control panel that the user scrolls past and never sees again. They will tune blindly, give up, or guess. Either keep the controls in view or duplicate a compact control bar next to each variation row.
When the artifact is a blog post, doc, copy deck, script, or media set, the user is not turning knobs — they're marking up the work the way an editor marks a manuscript. The studio renders the real artifact WYSIWYG (the actual rendered blog with its real components/media, not a raw-markdown textarea) and lets the user act on it directly. Building this for a doc review is mandatory: do not paste a long file into chat and ask "what do you think?" — that's the boring wall of text the user is rejecting. Stand up the