laniameda-kb（拉尼亚梅达 kb）

laniameda-kb

Save content to the laniameda.gallery Convex KB via the HTTP API.

Project

• Name: laniameda.gallery
• Repo: /Users/michael/work/laniameda/laniameda.gallery

Script

bun run ~/.agents/skills/laniameda-kb/scripts/ingest.ts '<json>'

Pillar auto-classification (your job as agent)

Content is organized into 4 pillars. Always determine the pillar before saving. Use the rules below.

When in doubt: dump. Never leave pillar empty.

Quick classification examples

• "cinematic fashion shoot, editorial lighting, high contrast" → creators
• "BMW M3 on mountain road, volumetric fog, 8K" → cars
• "SaaS landing page dark theme, glassmorphism" → designs
• "sunset over ocean, watercolor style" → dump
• portrait prompt with people → creators
• UI screenshot or web design reference → designs

Key fields

• ownerUserId — read from KB_OWNER_USER_ID env var automatically, never pass it
• pillar — always set this (see auto-classification above): "creators" | "cars" | "designs" | "dump"
• promptText — text/prompt content
• tagNames — array of tags (always include category + pillar as tag)
• imagePath — local path for inbound media (preferred over base64)
• imageUrl — remote URL to fetch and store
• ingestKey — auto-generated if omitted (dedup protection)
• modelName — AI model used: "Midjourney", "FLUX", "Nano Banana Pro", "Runway", "Kling", "Sora", "CDANCe", etc.
• generationType — "image_gen" | "video_gen" | "ui_design" | "other"
• promptType — "image_gen" | "video_gen" | "ui_design" | "cinematic" | "ugc_ad" | "other"
• domain — freeform category: "fashion", "portrait", "automotive", "architecture", "web design", etc.

Tagging

Always include a category tag: prompts, tutorials, resources, or ideas. Always include the pillar as a tag: creators, cars, designs, or dump. Add relevant style/tool tags: cinematic, midjourney, flux, fashion, portrait, automotive, ui, etc.

Common patterns

Single image + prompt (most common — Telegram forward):

bun run ~/.agents/skills/laniameda-kb/scripts/ingest.ts '{
  "promptText": "Cinematic close-up, golden hour, 35mm film grain, fashion editorial...",
  "imagePath": "/Users/michael/.openclaw/media/inbound/file_xxx.jpg",
  "pillar": "creators",
  "modelName": "Midjourney",
  "generationType": "image_gen",
  "domain": "fashion",
  "tagNames": ["prompts", "cinematic", "midjourney", "creators"]
}'

Image only (no prompt text — reference save):

bun run ~/.agents/skills/laniameda-kb/scripts/ingest.ts '{
  "imagePath": "/Users/michael/.openclaw/media/inbound/file_xxx.jpg",
  "pillar": "designs",
  "domain": "web design",
  "tagNames": ["resources", "ui", "designs"]
}'

One prompt + multiple variation images: Use the same promptText + promptIngestKey in every item:

bun run ~/.agents/skills/laniameda-kb/scripts/ingest.ts '[
  { "promptText": "...", "promptIngestKey": "my-slug", "pillar": "cars", "imagePath": "/path/1.jpg", "tagNames": ["prompts", "cars"] },
  { "promptText": "...", "promptIngestKey": "my-slug", "pillar": "cars", "imagePath": "/path/2.jpg", "tagNames": ["prompts", "cars"] }
]'

Batch (distinct unrelated items):

bun run ~/.agents/skills/laniameda-kb/scripts/ingest.ts '[
  { "promptText": "...", "pillar": "creators", "tagNames": ["prompts", "creators"] },
  { "imageUrl": "https://...", "pillar": "designs", "tagNames": ["resources", "designs"] }
]'

⚠️ Orphaned assets mistake: If you save variation images WITHOUT promptText + promptIngestKey, assets are saved as orphans. Always include both when using the variations pattern.

After saving

Report: pillar assigned, what was saved, model tagged, tags applied. If duplicate (same ingestKey) → "already in KB".

Schema changes

If the Convex schema has changed, read references/convex-interface.md before running.