laniameda-storage（拉尼亚梅达存储）

laniameda-storage

Save content to the Convex prompt-storager KB via the HTTP API.

Script

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

Key fields

• ownerUserId — read from KB_OWNER_USER_ID env var automatically, never pass it in JSON
• promptText — text/prompt content
• tagNames — array of tags (see tagging below)
• 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 that generated this (e.g. "Midjourney", "FLUX", "Nano Banana Pro", "Runway", "Kling", "Sora", "CDANCe")
• generationType — "image_gen" | "video_gen" | "ui_design" | "other"
• promptType — "image_gen" | "video_gen" | "ui_design" | "cinematic" | "ugc_ad" | "other"
• domain — freeform category: "product photography", "portrait", "architecture", "fashion", etc.

Single prompt + image with model tag:

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

Tagging

Always include a category tag: prompts, tutorials, resources, or ideas.

Keep tags digestible (default): 3–5 tags total.

• Don’t dump long keyword lists.
• Don’t include generation parameters as tags (e.g., Midjourney flags like --ar, --raw, etc.).

If the user explicitly asks for creative tags, include the literal tag creative (and still keep the total to ~3–5).

Add relevant tool/style tags as needed: midjourney, flux, cinematic, portrait, character, etc.

Common patterns

Single prompt:

bun run ~/.agents/skills/laniameda-storage/scripts/ingest.ts '{
  "promptText": "...",
  "tagNames": ["prompts", "tag1"]
}'

Single prompt + image (inbound Telegram media):

bun run ~/.agents/skills/laniameda-storage/scripts/ingest.ts '{
  "promptText": "...",
  "imagePath": "/Users/michael/.openclaw/media/inbound/file_xxx.jpg",
  "tagNames": ["prompts", "ai-image"]
}'

One prompt with multiple variation images (e.g. same template, different brand examples): Use the same promptText + same promptIngestKey in every item. The prompt is created once (deduped on promptIngestKey); each asset is linked to the same prompt.

bun run ~/.agents/skills/laniameda-storage/scripts/ingest.ts '[
  { "promptText": "...", "promptIngestKey": "my-prompt-slug", "imagePath": "/path/example1.jpg", "tagNames": ["prompts", "fashion"] },
  { "promptText": "...", "promptIngestKey": "my-prompt-slug", "imagePath": "/path/example2.jpg", "tagNames": ["prompts", "fashion"] },
  { "promptText": "...", "promptIngestKey": "my-prompt-slug", "imagePath": "/path/example3.jpg", "tagNames": ["prompts", "fashion"] }
]'

Result: ONE prompt record, THREE assets — all linked to the same prompt via promptId.

When to use this pattern:

• User shares a prompt template with multiple example outputs/variations
• Multiple brand applications of the same prompt (e.g. fashion lookbook for Ozon, Yandex, etc.)
• Same prompt, different style results

Batch (multiple distinct prompts or unrelated images): Pass a JSON array without a shared promptIngestKey — each item is fully independent:

bun run ~/.agents/skills/laniameda-storage/scripts/ingest.ts '[
  { "promptText": "prompt one", "tagNames": ["prompts", "cinematic"] },
  { "promptText": "prompt two", "imagePath": "/path/img.jpg", "tagNames": ["prompts", "portrait"] },
  { "imageUrl": "https://...", "tagNames": ["resources"] }
]'

Returns array of {promptId?, assetId?} — one result per item.

When to use batch vs. prompt+variations:

• Multiple distinct prompts → plain batch (no shared promptIngestKey)
• Same prompt + multiple example images → shared promptIngestKey pattern above
• Never concatenate multiple distinct prompts into one promptText

⚠️ Common mistake — orphaned assets: If you save variation images WITHOUT promptText + promptIngestKey, those assets will be saved as orphans (no prompt link). Always include both in every item when using the variations pattern.

After saving

Report: what was saved, tags applied. If duplicate (same ingestKey), say "already in KB".

Schema changes

If the Convex schema or action signature has changed, read references/convex-interface.md before running — it has the source file paths and current interface docs.