Wine Archive

Store, recall, and manage personal wine tastings and labels using natural language queries with durable image storage in a local SQLite database.

Personal wine archive for tasting history, label storage, and natural-language recall. Stores wines in a local SQLite database with a normalized schema: stable wine identities plus individual tasting/purchase instances. Integrates with Telegram via the OpenClaw bridge.

Trigger phrases

Use this skill when the user says things like:

• "remember this wine", "log this bottle", "add a wine", "save this tasting"
• "what wine did I have last week?", "show me my pinot noirs", "recall wines from Nugget"
• "show the [wine name] label", "send me the Vinho Verde photo"
• "show me the [wine name]", "tell me about the Broadbent"
• "export my wine archive", "import wine data"

Setup (first time)

bash

cd skills/wine-archive
zsh setup.sh

Or manually:

bash

cd skills/wine-archive
npm install
npm run wine:init
cp .env.example .env   # add ANTHROPIC_API_KEY for LLM features

Data is stored in data/wine/wine.sqlite3 and label images in data/wine/labels/. Override the DB location with WINE_DB_PATH env var.

Core rules

• Personal ratings use a 5-star scale.
• Prefer new tasting instances over overwriting when the user re-tries a wine.
• If it is ambiguous whether to update an existing instance or create a new one, ask.
• Keep deterministic regex parsing as the source of truth; LLM parsing is assistive only.
• Archived label images are durable — never rely on temporary inbound media paths.
• On Telegram, use the bridge path for label recall rather than inline MEDIA: text.

Data model

Wine = stable identity

• producer, wine_name, region, country, style, color, varietal
• official_rating / official_rating_source
• label image (default_source_image_path)

Wine instance = specific bottle / tasting event

• vintage, price, currency, place_of_purchase
• purchased_on, consumed_on
• subjective_rating (5-star), notes
• source_type (chat | image | manual)

Commands

Initialize DB:

bash

npm run wine:init

Add from free text:

bash

npm run wine:add -- --text "Had a Broadbent Vinho Verde from Minho. Bought at Nugget for $14. Rated 4/5."

Add structured:

bash

npm run wine:add -- --wine_name "Broadbent Vinho Verde" --varietal "Loureiro" --region "Minho" \
  --style "vino verde" --color "white" --price 14 --place_of_purchase "Nugget" --consumed_on 2026-03-29

Add from label text + image:

bash

npm run wine:add -- --label-text $'Broadbent\nVinho Verde\n2024\nMinho Portugal\nLoureiro' \
  --image data/wine/labels/broadbent.jpg

Parse label without inserting:

bash

npm run wine:parse-label -- --label-text $'Broadbent\nVinho Verde\n2024\nMinho Portugal'

Query:

bash

npm run wine:query -- --text "vinho verde"
npm run wine:query -- --varietal "Pinot Noir"
npm run wine:query -- --consumed_after 2026-03-24 --consumed_before 2026-03-31

Natural-language recall:

bash

npm run wine:recall -- --text "show me wines from last week at Nugget"
npm run wine:recall -- --text "what pinot noir did I drink last month"
npm run wine:recall -- --text "find red wines rated at least 3"

Chat-facing flow (intent detection + dispatch):

bash

npm run wine:chat -- --text "Remember this wine: Had a vinho verde last week. Rated 4/5."
npm run wine:chat -- --text "What was that vinho verde I had last week?"
npm run wine:chat -- --label-text $'Broadbent\nVinho Verde\n2024\nMinho' --image data/wine/labels/broadbent.jpg

List recent entries:

bash

npm run wine:list -- --limit 10

Remove an entry:

bash

npm run wine:remove -- --id 42

Export archive:

bash

npm run wine:export -- --out my-wines.json              # paths only
npm run wine:export -- --out my-wines.json --include-images  # embed label images as base64

Import archive:

bash

npm run wine:import -- --in my-wines.json --dry-run    # preview only
npm run wine:import -- --in my-wines.json               # import

Telegram bridge (requires OpenClaw)

The bridge script resolves the wine chat request and outputs a structured JSON result. When action === "send-media", the agent sends the media using the openclaw CLI.

bash

npm run wine:telegram-bridge -- --text "Show me the Vinho Verde label"

Example output when a label is found:

json

{
  "status": "ok",
  "action": "send-media",
  "reply": "Broadbent Vinho Verde label",
  "mediaPath": "./data/wine/labels/broadbent-abc123.jpg",
  "caption": "Vinho Verde label"
}

The agent then sends:

bash

openclaw message send --channel telegram \
  --target <chat_id> --thread-id <thread_id> --reply-to <reply_to> \
  --media <mediaPath> --message <caption>

Shell helper (outputs JSON for the agent):

bash

zsh scripts/wine-send-label-telegram.sh <chat_id> <thread_id> <reply_to> Vinho Verde

Response templates

Add / save confirmation

text

Added as a new wine instance:
- Broadbent Vinho Verde · 2024
- Nugget, $14.00 · consumed 2026-03-29
- rated 4/5

Update confirmation

text

Updated — rating: 3.5/5, consumed_on: 2026-04-02

Show a wine (canonical entry)

text

Broadbent — Vinho Verde
- region: Minho
- country: Portugal
- style: vino verde
- color: white
- varietal: Loureiro
[label image]

Show wine instances

text

Broadbent — Vinho Verde (2024)
- consumed: 2026-03-29
- rated: 4/5
- notes: crisp, slightly frizzante

Show a list of wines

text

- Vinho Verde
  producer: Broadbent
  year: 2024
  varietal: Loureiro
  style: vino verde

Recall by time window

text

Broadbent — Vinho Verde (2024)
- consumed: 2026-03-29
- rated: 4/5

Label response

Send the archived image with caption: Vinho Verde label

Clarification (ambiguous intent)

text

Do you want me to update the existing instance or create a new tasting instance?

Environment variables

Variable	Default	Description
ANTHROPIC_API_KEY	—	Required when using an Anthropic model for LLM intent classification
OPENAI_API_KEY	—	Required when using an OpenAI model for LLM intent classification
WINE_DB_PATH	data/wine/wine.sqlite3	Override DB location
WINE_LLM_INTENT_CLASSIFIER	0	Set to 1 to enable LLM intent classification (off by default; skill works fully offline without it)
WINE_LLM_INTENT_MODEL	openai/gpt-4.1-mini	Model for intent classification — use claude-haiku-4 for Anthropic, openai/gpt-4.1-mini for OpenAI
WINE_LLM_INTENT_MIN_CONFIDENCE	0.65	Minimum confidence to use LLM over regex

Platform notes

• macOS: Full support including image normalization via sips.
• Linux/Windows: Core features work. Image resizing is skipped (install ImageMagick and adapt lib/wine-store.js normalizeImageInPlace for cross-platform resizing).
• Node.js >= 22 required (node:sqlite built-in used in shared/interaction-store.js).