Auto

View saved version →

---
name: auto-photos
description: Search and display photos from the user's Auto camera-roll library through the Auto MCP plugin. Use when the user explicitly asks to find, browse, show, or view "my photos," "my pictures," or otherwise clearly refers to their own photo library, including searches by subject, visible detail, scene, activity, mood, date or date range, location, or a combination of these. Also use when Auto or this skill is explicitly invoked, and for follow-ups referring to a photo or photo set previously returned by Auto. Do not use for generic requests such as "show me pictures of New York" that do not indicate the user's own photos; those should use ordinary image search instead.
---

# Auto Photos

Search the user's Auto library and render the results directly in the conversation. Keep the interaction visual and concise.

## Search

1. Parse the request into:
   - semantic content such as people, objects, clothing, text, scenes, activities, colors, or mood;
   - structured filters such as date range, tags, or location;
   - ordering requested or implied by the user.
2. Call `auto_search_photos` for semantic searches or composed filters. Put content and mood in `query`; pass time separately with `startDate` and `endDate`. Use ISO 8601 UTC timestamps and interpret relative dates in the user's timezone. Location is provided as Lat/Long and radius in meters.
3. Call `auto_list_photos` for plain chronological browsing without a semantic query.
4. Treat mood and visual-feel searches as best-effort semantic matches. Do not claim that every result definitively has the requested mood.
5. Prefer one good search over several speculative searches. If the request is genuinely ambiguous and the alternatives would produce materially different photos, ask one short clarifying question.

Photo descriptions are detailed visual embeddings. Phrase semantic queries around visible evidence rather than metadata the description is unlikely to contain. Combine subject and distinctive details when useful, but do not invent details the user did not request.

## Choose the presentation

Select the view without asking how many photos to show:

- Use `auto_display_photo` for a singular request, a specific known photo, or the strongest single answer.
- Use `auto_display_photos_carousel` for practical, ordered results such as recent photos, a date range, a sequence, progress over time, or “show me some.” Display 5–10 photos; default to 7 when no better count is implied. Preserve the useful order, normally newest first unless the user requests otherwise.
- Use `auto_display_photos_grid` for unordered visual exploration, categories, broad subjects, aesthetics, or a “feel” or mood. Display 9–12 photos; default to 12 when enough good matches exist.

If fewer suitable results exist, show the available results. Never duplicate or pad results. Honor an explicit user count when practical, even if it differs from the defaults.

## Display results

After searching, immediately call the appropriate Auto display tool with the returned photo IDs. Do not substitute Markdown links, manually embedded images, or a prose list for the Auto viewer. Keep any introduction to one short sentence, and do not narrate every image unless the user asks.

If nothing matches, say so plainly and offer one useful broader search. Do not display unrelated low-confidence results merely to fill the view.

## Follow-ups and inspection

Resolve phrases such as “the third one,” “that photo,” or “show that one” against the most recently displayed set and its order. Use `auto_display_photo` to enlarge a selected result.

Use `auto_get_photo` when metadata or memory-derived descriptions are needed. Use `auto_get_photo_image` only when the user asks for analysis requiring the actual pixels; it buffers the full image and should be called sparingly.

Never use web image search or image generation for requests about the user's own Auto photos.