lov-find-logo

find-logo — collect brand logos, prefer wide + transparent

Takes a brand name or URL, probes Clearbit + the site's own og:image / <link rel=icon> / favicon, scores each candidate, and archives the best one plus a couple of alternates into the configured collection directory.

When to Use

• User asks to collect one or more brand logos for a slide/poster/site lineup
• User names companies to drop into a partners/press strip
• User gives a URL and wants its logo pulled down cleanly

Workflow (MANDATORY)

Step 1: Identify each brand

Accept any mix of names and URLs. If the user gave only a name with no obvious domain, ask — don't silently guess .com (script will guess, but for non-US or ambiguous brands that fails).

Use AskUserQuestion when:

• Brand name is ambiguous (e.g. "Apple" = fruit vs. Inc.)
• No URL and the domain isn't guessable (xAI → x.ai, not xai.com)
• User gave a list without URLs

Step 2: Fetch — one brand per invocation

python3 scripts/find_logo.py --name "Anthropic" --url https://anthropic.com --json

For a batch, loop; the script is idempotent per <slug>/ (re-runs overwrite).

Step 3: Inspect score; fall back to WebSearch if needed

• Exit code 0 → logo archived. The printed score is your quality signal:
  • ≥ 60 — solid: SVG or transparent PNG with wide/square aspect
  • 20–60 — usable: probably a favicon or small PNG
  • < 20 — weak: only ICO or tiny stub found
• Exit code 2 / status: "no-candidates" → script found nothing. Do NOT give up. Use WebSearch for "<brand> logo svg site:*.com" or the brand's press-kit page, then re-invoke with --url <direct-image-url> is not supported — if you have a direct image URL, save it into the configured collection directory under <slug>/logo.<ext> and hand-write meta.json using the existing layout as a template.

Step 4: Report

Report back with the archive path and the primary's aspect + format. If the score is weak, tell the user and offer to retry with a specific press-kit URL or Wikipedia SVG.

CLI Reference

At least one of --name or --url is required.

Archive Layout

~/.skill-publisher/logo-collection/
├── anthropic/
│   ├── logo.png            # primary (highest score)
│   ├── alt-1.png           # runner-ups
│   ├── alt-2.png
│   └── meta.json           # sources, scores, dimensions, fetched_at
├── vercel/
│   ├── logo.png            # 1200x628 transparent banner
│   └── ...
└── stripe/
    ├── logo.svg
    └── ...

Scoring Heuristic (why a candidate wins)

• Format: SVG (+40) > PNG (+20) > WebP (+10) > JPG (-10) > ICO (-20)
• Transparency: +30 if alpha channel present (SVG always counts)
• Aspect ratio: +25 for wide (≥2:1), +10 for landscape (≥1.3:1), -5 for square, -15 for tall/portrait
• Short edge: +15 if ≥128px, +5 if ≥64px, -20 if <32px
• Size sanity: -30 if payload <400 bytes (almost certainly a stub)

This matches the "prefer 长条形 + rgba" preference — wide transparent logos come out on top, square favicons land as alternates.

Dependencies

Stdlib only (urllib, html.parser, argparse). No pip install required.

User Configuration

Default archive files live under ~/.skill-publisher/logo-collection/. Override this per run with --out, or set SKILL_FIND_LOGO_OUTPUT_DIR for the skill.

Known Limits

• The name → domain guess is a crude lowercase-strip + .com suffix. For anything not on .com, pass --url explicitly.
• No Clearbit API key is used — we hit the unauthenticated endpoint, which covers most major brands but not all.
• WebSearch fallback is Claude's responsibility, not the script's.

Runtime context (shared)

运行前读取本 Skill 包的 skill.yaml，由宿主提供 skill-runtime/v1 上下文。字段解析顺序为：当前请求、项目上下文、个人 Preferences、品牌 Profile、通用默认值。

• 只使用 Manifest 声明的字段；Profile 保存公开品牌事实，Preferences 保存个人工作偏好。
• required: true 字段缺失时，按 Manifest 的问题配置向用户提出一个聚焦问题；用户明确同意后再保存回答。
• 报错提供可复制的 context_id、字段路径与来源，诊断内容避开秘密、完整私人路径和原始配置。

通用反馈闭环

用户在 Skill 驱动任务中提出修改意见时，继续当前产物前必须执行：

1. 先判断意见是 task-specific（仅本次）还是 reusable（可跨任务复用）。
2. task-specific 只修改当前任务，不改 Skill。
3. reusable 先确定作用域：领域规则先更新对应 canonical Skill；适用于所有 Skill 的规则先更新共享规范。
4. 完成规则更新、版本、lint 与分发核验后，再把修改应用到当前任务。
5. reusable 修改会使此前的“确认”“继续”“发吧”失效；完成当前产物修改和回读后必须停下，等待用户下一步指示，不自动进入发布、提交或其他外部写入。