michailbul/laniameda-gallery-query

laniameda-gallery-query

Use this skill to read from laniameda.gallery.

It covers two read surfaces:

• asset-centric reads: browse assets, semantic search, fetch one asset, download media
• pack reads: fetch a saved asset pack and its member assets from a copied gallery ID
• designs-pillar reads: browse structured design inspirations and inspect one design entry with its linked preview asset

Counterpart to laniameda-gallery-ingest (which writes).

Runtime env

• CONVEX_URL or NEXT_PUBLIC_CONVEX_URL — required
• KB_OWNER_USER_ID — required for owner-scoped reads (scope: "mine") and all designs-pillar actions

Best practice:

• keep KB_OWNER_USER_ID env-driven
• do not hardcode Michael's Telegram ID into wrappers or prompts
• use scope: "public" only for public asset discovery; design inspiration reads are owner-scoped

Script

cd ~/work/laniameda/laniameda.gallery
bun run skills/laniameda-gallery-query/scripts/query.ts '<JSON>'

Actions

list

Browse assets with structured filters.

Supported filters:

• scope: mine or public (mine default)
• pillar
• kind
• modelName
• folderId (mine only)
• assetRole
• search (text filter on hydrated asset content)
• limit

Example:

{
  "action": "list",
  "scope": "mine",
  "pillar": "cars",
  "assetRole": "reference",
  "folderId": "folders:abc123",
  "limit": 10
}

search

Semantic asset search via semanticSearch:searchAssets.

Supported filters:

• scope
• query
• pillar
• kind
• modelName
• folderId (mine only)
• assetRole
• limit

Example:

{
  "action": "search",
  "query": "dark moody editorial portrait with film grain",
  "scope": "mine",
  "pillar": "creators",
  "assetRole": "generated_output",
  "limit": 5
}

get

Fetch one owner-scoped asset with hydrated prompt/tag metadata.

{
  "action": "get",
  "assetId": "asset:abc123"
}

Raw Convex asset IDs are also accepted for assetId, but copied gallery IDs use the typed asset:<id> form.

getPack

Fetch one owner-scoped asset pack and its hydrated member assets.

{
  "action": "getPack",
  "packId": "pack:abc123"
}

getById

Resolve a copied gallery ID without first deciding which table to query.

Supported copied ID formats:

• asset:<id>
• pack:<id>
• design:<id>

These tokens are produced by the gallery UI when the user clicks:

• The corner copy button on any asset card (hover on desktop)
• The persistent asset:<id> chip in the detail panel metadata strip
• "Copy asset / design / pack ID" items inside the detail panel Copy dropdown

Accept the pasted token verbatim — do not strip the prefix. Raw Convex IDs are also accepted, but the typed form lets the skill resolve the correct table automatically.

Example:

{
  "action": "getById",
  "id": "pack:abc123"
}

download

Download one owner-scoped asset to local disk.

{
  "action": "download",
  "assetId": "asset:abc123",
  "outDir": "/tmp/laniameda-gallery"
}

listDesigns

Browse structured entries from the designInspirations pillar.

Supported filters:

• inspirationType
• platform
• workflowType
• captureKind
• saveIntent
• folderId
• sourceDomain
• search
• dateFrom
• dateTo
• requireAsset
• limit

Example:

{
  "action": "listDesigns",
  "platform": "web",
  "captureKind": "website",
  "saveIntent": "inspiration",
  "search": "pricing",
  "requireAsset": true,
  "limit": 10
}

getDesign

Fetch one owner-scoped design inspiration and, when present, hydrate its linked preview asset.

{
  "action": "getDesign",
  "designInspirationId": "design:abc123"
}

Typical workflows

Find and reuse an image prompt

1. search to find the best asset
2. download to save the asset locally
3. use savedPath and promptText in the current task

Resolve a copied gallery item

1. Use getById with the exact copied ID from the gallery UI
2. If the ID starts with pack:, inspect the returned assets array and choose the needed member asset
3. If media bytes are needed, run download with the chosen asset:<id>

Find a saved design reference

1. listDesigns with search, platform, or captureKind
2. getDesign to inspect the full record and linked preview asset

Response highlights

Asset actions return compact asset objects with fields like:

• id
• kind
• pillar
• modelName
• promptText
• tagNames
• url
• thumbUrl
• folderId
• assetRole
• assetPackId
• packSlotIndex
• score (semantic search only)

Pack actions return:

• pack.id
• pack.title
• pack.description
• pack.pillar
• pack.modelName
• pack.coverAssetId
• pack.itemCount
• assets hydrated like asset results

Design actions return compact design objects with fields like:

• id
• title
• summary
• sourceUrl
• sourceTitle
• userNote
• inspirationType
• platform
• workflowType
• captureKind
• saveIntent
• templateKey
• sourceFingerprint
• previewUrl
• previewThumbUrl
• assetId
• promptId

Notes

• Semantic asset search requires SEMANTIC_EMBEDDINGS_ENABLED=true on the Convex deployment.
• The embedding model in this repo is gemini-embedding-2-preview.
• download saves raw bytes; video assets are not transcoded.
• Convex storage URLs are temporary. Download promptly after retrieval.