Lov Maintain Partners

maintain-partners — Skill Publisher 合作伙伴板块维护

Maintains the configured website repo. Resolve the path from --repo, SKILL_MAINTAIN_PARTNERS_SITE_ROOT, or the shared user profile. The partners strip usually lives in app/(main)/(home)/PartnersGrid.tsx as a PARTNERS: Partner[] array; older sites may still keep it in app/(main)/(home)/WorkshopDispatch.tsx. Logos live in public/partners/<slug>/logo.png; taglines in src/i18n/messages/{zh-CN,en,ja,th}.json under dispatch.partner*Tagline.

User Configuration

Before touching files, resolve:

SKILL_ROOT="${SKILL_SKILLS_INSTALL_DIR:?Set SKILL_SKILLS_INSTALL_DIR}"
SKILL_DIR="${SKILL_DIR:-$SKILL_ROOT/lov-maintain-partners}"
WEB_ROOT="${SKILL_MAINTAIN_PARTNERS_SITE_ROOT:?Set this or pass --repo}"
PARTNERS_TSX="${SKILL_MAINTAIN_PARTNERS_FILE:-app/(main)/(home)/PartnersGrid.tsx}"

Use this precedence for the website root:

1. Explicit --repo <path> on add_partner.py / audit_partners.py.
2. SKILL_MAINTAIN_PARTNERS_SITE_ROOT.
3. Shared profile JSON at ${SKILL_PROFILE_PATH:-$HOME/.skill-publisher/skills/profile.json}.

SKILL_WEB_ROOT and PARTNERS_SITE_ROOT are accepted as legacy aliases, but should not be the public contract for reusable skills.

Use this precedence for the partners TSX file:

1. Explicit --partners-file <path>.
2. SKILL_MAINTAIN_PARTNERS_FILE.
3. Shared profile keys sites.partners_file, skill-publisher.partners_file, partners.file, or workspace.partners_file.
4. app/(main)/(home)/PartnersGrid.tsx, then legacy app/(main)/(home)/WorkshopDispatch.tsx.

SKILL_PARTNERS_FILE and PARTNERS_FILE are accepted as legacy aliases, but should not be the public contract for reusable skills.

For details and supported profile keys, read references/user-config.md.

Skill Dependencies

• lov-find-logo is required for all logo discovery. This skill must not scrape homepages itself or keep a separate fallback crawler.
• Use the depends_on frontmatter field to declare skill-level dependencies. This mirrors the depends_on field in lov-general-skills/skills.yaml; unknown frontmatter keys are expected to be ignored by agents that do not consume dependency metadata.

When to Use

• User asks to add one or more new partners (with or without a logo URL).
• User asks to standardize / normalize a logo (sizing wrong, white-on-white, etc.).
• User provides a local file and asks to replace an existing partner's logo.
• User asks to audit the partners section before a release.

Standards

• Logo canvas: 80px content height for the website partners strip (light grayscale, CSS height: 32px ≈ 2.5× density, sharp enough), 240px for event posters or any retina export at scale: 2 or higher.
• For white-on-transparent logos: invert (full or selective) so they show on the light grayscale strip.
• For icon-only logos < ~40px wide after normalization: pass --show-name when adding so the brand name renders next to the icon.
• Tagline format: <品牌名> · <一句话定位> in Chinese; mirror style in en/ja/th.

Workflow

Op 1: Add a new partner

1. Ask the user for the brand name + homepage URL via AskUserQuestion.
2. Collect the logo with lov-find-logo:

   python3 "$SKILL_ROOT/lov-find-logo/scripts/find_logo.py" \
     --name "<显示名>" --url <URL> --slug <slug> --json
   

   Use the archived primary asset under ~/.skill-publisher/logo-collection/<slug>/logo.<ext>. If find_logo.py returns no candidates, stop and ask the user for a better official URL / press-kit URL, then rerun find_logo.py. Do not call a local scraper from this skill.
3. Visually verify the archived primary asset before normalizing.
4. If the primary asset is SVG, rasterize it before normalization:

   rsvg-convert -h 240 ~/.skill-publisher/logo-collection/<slug>/logo.svg \
     -o /tmp/<slug>-raw.png
   

   Use the rasterized /tmp/<slug>-raw.png as --src. For non-SVG sources, use the archived primary asset directly.
5. Normalize:

   python3 "$SKILL_DIR/scripts/normalize_logo.py" \
     --src <archived-or-rasterized-logo> \
     --dst "$WEB_ROOT/public/partners/<slug>/logo.png" \
     --invert auto
   

6. Read the normalized PNG to confirm it's visible (not white-on-white).
7. Append to PARTNERS + all 4 locale JSONs:

   python3 "$SKILL_DIR/scripts/add_partner.py" \
     --repo "$WEB_ROOT" \
     --partners-file "$PARTNERS_TSX" \
     --name "<显示名>" --href "<URL>" \
     --logo "/partners/<slug>/logo.png" \
     --key partner<Slug>Tagline \
     --category community \
     --zh "..." --en "..." --ja "..." --th "..." \
     [--show-name]
   

Op 2: Normalize an existing logo

python3 "$SKILL_DIR/scripts/normalize_logo.py" \
  --src public/partners/<slug>/logo.png \
  --dst public/partners/<slug>/logo.png \
  --invert auto

Re-read after to verify.

Op 3: Replace logo from a user-provided file

Ask for the source file path directly, or read it from the user's configured workspace/profile. Do not assume a private partners folder.

python3 "$SKILL_DIR/scripts/normalize_logo.py" \
  --src "<user-provided path>" \
  --dst "$WEB_ROOT/public/partners/<slug>/logo.png" \
  --invert auto

JPEG inputs auto-strip near-white background to transparent before crop.

Op 4: Audit

python3 "$SKILL_DIR/scripts/audit_partners.py" \
  --repo "$WEB_ROOT" \
  --partners-file "$PARTNERS_TSX"
# add --probe to also HTTP-check every href (slow, requires proxy)

Reports: missing logo files, missing i18n keys per locale, dead URLs.

Op 5: Align a row of partner logos (cross-asset visual height parity)

When: putting 3+ partner logos in a single horizontal strip and they look different sizes despite having the same CSS height. Common in event posters, hero sections, "联办 / co-host" rows.

Root cause: each source file has different internal padding (designer canvas margin), so two PNGs both set to height: 24px render at different visible heights because their content occupies different fractions of the canvas. Per-logo CSS height tweaks based on eyeballed content ratios are unstable—different displays / scaling will diverge again.

Reliable fix — trim at file level, uniform CSS box:

1. Normalize every logo to identical content height. Default raster file target is 240px (3× density for retina poster export at scale: 2; 80px gives only 1.7× and looks soft after PNG export). Use --invert off if the source is already light-on-transparent (don't double-invert):

   for f in lujiazui juanyi citic-bookstore citic-thinker-lab; do
     python3 "$SKILL_DIR/scripts/normalize_logo.py" \
       --src "<configured-partners-source>/<brand>/<file>.png" \
       --dst <event-assets>/partners/$f.png \
       --height 240 --invert auto
   done
   

   Always normalize from the original source, never from a previously normalized 80px file (upscaling = blurry — burned by this on juanyi).

2. For SVG sources, rasterize first. normalize_logo.py operates on raster pixels and cannot crop SVG viewBox padding. Without this step an SVG always renders smaller than rasterized PNG siblings:

   rsvg-convert -h 720 brand.svg -o /tmp/brand-raw.png   # 3× of 240
   python3 "$SKILL_DIR/scripts/normalize_logo.py" \
     --src /tmp/brand-raw.png --dst <event-assets>/partners/brand.png \
     --height 240 --invert off
   

   rsvg-convert ships with librsvg (brew install librsvg).

3. For SVG with embedded background rect (icon wrapped in a black/colored rounded square — common in app-icon-style SVGs from find-logo), strip the background before rasterizing, otherwise filter brightness(0) invert(1) flattens it into a solid white block that hides the icon:

   # Drop the outer <rect fill="#000"...> wrapper
   sed -E 's|<rect[^/]*fill="#0+"[^/]*/>||' brand.svg > /tmp/brand-clean.svg
   rsvg-convert -h 720 /tmp/brand-clean.svg -o /tmp/brand-raw.png
   

4. Wrap each logo in a fixed-size box (recommended over auto-width flex):

   <span class="ps-logo-box"><img src="..." class="ps-logo"></span>
   

   .ps-logo-box {
     width: 96px; height: 30px;             /* fixed grid cell */
     display: inline-flex;
     align-items: center; justify-content: center;
     border: 1px solid rgba(255,255,255,0.10);
     border-radius: 4px;
     padding: 3px 6px;
     box-sizing: border-box;
   }
   .ps-logo { max-width: 100%; max-height: 100%; width: auto; height: auto; display: block; }
   

   Fixed boxes give a stable matrix look — narrow logos (icon-only) and wide logos (icon + wordmark) all occupy the same footprint, with the asset scaled to fit. Auto-width flex (the older recipe) makes per-row total widths unpredictable as logos get added/removed.

5. Dark-background unification — when the row sits on a dark canvas (e.g. event poster), most brand logos are designed for white BG and look inconsistent (some have black text, some have brand-colored marks). The stable recipe:

   .ps-logo { filter: brightness(0) invert(1) opacity(0.88); }
   /* logos already white-on-transparent — opt out of inversion */
   .ps-logo.ps-logo-original { filter: opacity(0.88); }
   

   brightness(0) flattens all colors to black, then invert(1) produces uniform white at the configured opacity. The .ps-logo-original escape hatch is for source files that are already white-on-transparent (white SVG variants from a brand kit) so you don't double-process them into invisible black-on-dark.

6. Icon-only SVG → composite icon + wordmark — if the brand SVG only has an icon (no "BrandName" wordmark beside it), don't ship just the icon in a 96×30 box (it'll look like an unidentified mark). Compose the wordmark with PIL using the brand's own font when possible:

   from PIL import Image, ImageDraw, ImageFont, ImageOps
   # 1. rasterize cleaned SVG, invert white→black so default filter works
   icon = Image.open('/tmp/brand-icon.png').convert('RGBA')
   r, g, b, a = icon.split()
   inv = Image.merge('RGB', (ImageOps.invert(r), ImageOps.invert(g), ImageOps.invert(b)))
   icon = Image.merge('RGBA', (*inv.split(), a))
   icon = icon.crop(icon.getbbox())
   target_h = 240
   icon = icon.resize((int(icon.width * target_h / icon.height), target_h), Image.LANCZOS)
   # 2. render wordmark in brand font (find-logo bundles fonts/ when found)
   font = ImageFont.truetype('partners/<brand>/fonts/<Family>.ttf', 150)
   # 3. compose icon + gap + text on transparent canvas
   

   The PNG goes through the same brightness(0) invert(1) filter as raster logos — match colors with all other entries automatically. Use the brand's own font (often shipped under <brand>/fonts/ by the find-logo skill); fall back to system SF / Helvetica only if no brand font is available.

7. Anti-pattern — do not try to fix alignment by setting per-logo heights like .ps-logo-juanyi { height: 26px }. It's brittle (every new logo needs another magic number), unstable across browsers, and breaks the moment a designer reships the source asset with different padding.

CLI Reference

normalize_logo.py

add_partner.py

audit_partners.py

Network proxy

Sandbox child processes don't inherit the system ClashX proxy. Before fetching logos with lov-find-logo or probing partner URLs, export:

export https_proxy=http://127.0.0.1:7890 \
       http_proxy=http://127.0.0.1:7890 \
       all_proxy=socks5://127.0.0.1:7891

audit_partners.py already injects these for curl invocations.

Dependencies

git clone https://example.com/skills/find-logo-skill \
  "${SKILL_SKILLS_INSTALL_DIR:?Set SKILL_SKILLS_INSTALL_DIR}/lov-find-logo"
python3 -m pip install Pillow
brew install librsvg  # for SVG logo sources

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。