Create Shopping List (Gemini Flash + Todoist)

Target flow:

1. Input is either a photo or a recipe web search
2. Extract ingredients (Gemini Flash for photos; web_fetch text → Gemini for websites)
3. Pull current Todoist Shopping list
4. Compare using overlap + synonym mapping (kept conservative; only merge high-confidence equivalents like coriander↔cilantro, panko↔breadcrumbs)
5. Update Shopping (default: add only missing items; skip salt/pepper)

Use the bundled script to handle the photo → ingredients → Shopping update part.

It also automatically saves a markdown entry into recipes/ (your cookbook knowledge base) and appends to recipes/index.md.

For recipe-name → web search, do it confirm-first using web_search + web_fetch, then feed the ingredients into the same update logic (and save the recipe).

Prereqs

• Env: GEMINI_API_KEY (or GOOGLE_API_KEY) for Gemini
• Env: TODOIST_API_TOKEN for Todoist
• Bin: todoist (todoist-ts-cli)

Output formatting

• Items are reformatted to start with the ingredient name, followed by a parenthetical quantity.
• The Shopping list is kept flat (no Todoist sections/groups).

Run

python3 skills/recipe-to-list/scripts/recipe_to_list.py \
  --image /path/to/photo.jpg \
  --title "<optional title>" \
  --source "photo:/path/to/photo.jpg"

Optional flags

• --model gemini-2.0-flash (default; falls back automatically) or any compatible Gemini vision model
• --dry-run to print extracted items without creating tasks
• --prefix "[Recipe] " to prefix each created task
• --no-overlap-check to skip checking your existing Shopping list
• --include-pantry to include salt/pepper
• --no-save to skip saving into recipes/

What to send to the model

The script prompts Gemini to return strict JSON:

{
  "items": ["2 large globe eggplants", "kosher salt", "..."],
  "notes": "optional"
}

If parsing fails, rerun with a clearer crop (ingredients list only) or provide a manual list.