places - SKILL.md Agent Skill

name: places description: Save a place (restaurant, bar, activity, attraction) from a URL or description to the places database. Use when someone shares a place link, asks to save a spot, asks about saved places, or wants recommendations from saved places. metadata: {"clawdbot":{"emoji":"📍","requires":{"bins":["sqlite3"]}}}

Places → SQLite

Saving a place from a URL

Step 1: Fetch and extract place info

Instagram URLs (instagram.com/reel/<ID>/, instagram.com/p/<ID>/):

Extract the post ID from the URL first. The ID is the path segment after /reel/ or /p/. Remember it — you'll use it to verify you got the right content.
Try oEmbed first. It's scoped to the exact URL — no feed-scroll mistake possible:
```
curl -sf "https://api.instagram.com/oembed/?url=$URL" | jq '{title, author_name, author_url, thumbnail_url}'
```
Returns caption (title), creator handle (author_name), and profile URL (author_url). If this works, use it directly. Skip to Step 2 of the overall flow.
Verify the oEmbed match. The author_url must contain the handle from the input URL's path (e.g., URL was instagram.com/<handle>/reel/<ID>/ or the post links back to a profile consistent with the reel). If the handles diverge, discard oEmbed's result and fall through to the browser path.
Browser fallback (oEmbed returned an error, auth wall, rate limit, or mismatched handle):
- openclaw browser navigate "URL" — uses the default openclaw browser profile (logged in). NEVER pass profile: "user" — it will fail.
- Wait for page load.
- openclaw browser snapshot --format ai to extract text.
- Verify the snapshot matches the URL. The post ID you extracted in Step 1 must appear in the snapshot output (URL bar, meta, or embedded link). If it doesn't, Instagram's feed scrolled and you're looking at a different reel — navigate to the URL again and re-snapshot. Do NOT proceed with content from a different post.
- Click "See more" if needed to expand the full description.
Extract: place name, location, what kind of place it is, any pricing info.
Do NOT save anything if you couldn't verify the extracted content against the post ID from Step 1. Reply asking what the place is instead of guessing.

Facebook / TikTok URLs: Use the browser fallback path above (Steps 4-6). oEmbed is Instagram-specific here.

Do NOT use web_fetch for social media — it returns a login wall with no content.

Google Maps URLs (maps.google.com, goo.gl/maps, maps.app.goo.gl):

Use web_fetch or browser tool
Extract: name, address, category, price range, rating from the page

Review sites (yelp.com, tripadvisor.com, eater.com, infatuation.com):

Use web_fetch to get the page
Extract: name, location, cuisine/type, price range, highlights

All other URLs:

Use web_fetch to get the page
Extract whatever place info is available

Step 2: Save to places database

sqlite3 ~/places/places.sqlite "INSERT INTO places (name, category, source_url, source_platform, location, cuisine_or_type, tags, price_range, notes, saved_by) VALUES (
  'Place Name',
  'restaurant',
  'https://original-url.com',
  'instagram',
  'Queen West, Toronto',
  'Italian',
  '[\"date-night\", \"outdoor-patio\", \"pasta\"]',
  '\$\$',
  'Wife saw this on Instagram. Known for handmade pasta.',
  'wife'
);"

Adjust the database path to wherever you store your places.sqlite.

Field notes:

category: restaurant, cafe, bar, brewery, activity, park, attraction, shop, hotel, bakery
tags: JSON array — include vibe, occasion, features (e.g., "date-night", "kid-friendly", "outdoor", "brunch", "group-friendly")
source_platform: instagram, google, yelp, tiktok, website, tripadvisor, eater, manual
cuisine_or_type: for restaurants use cuisine ("Italian", "Japanese"); for activities use type ("climbing gym", "hiking trail", "escape room")
location: neighborhood + city, or full address if available
price_range: $, $$, $$$, $$$$ (leave NULL if unknown)
saved_by: who shared/requested it (manan, wife, etc.)
notes: context about why it was saved, who recommended it, what it's known for

Step 3: Saving a place from description (no URL)

When someone says "remember that sushi place in Kensington Market" or "save Blue Barn as a brunch spot":

Ask for any missing critical info (name and rough location at minimum)
Set source_platform to manual
Set source_url to NULL

Querying places

# Recent saves
sqlite3 ~/places/places.sqlite "SELECT id, name, category, location, cuisine_or_type, created_at FROM places ORDER BY created_at DESC LIMIT 10;"

# Search by name
sqlite3 ~/places/places.sqlite "SELECT id, name, category, location, cuisine_or_type, price_range FROM places WHERE name LIKE '%blue%';"

# Search by category
sqlite3 ~/places/places.sqlite "SELECT id, name, location, cuisine_or_type, price_range FROM places WHERE category = 'restaurant' ORDER BY created_at DESC;"

# Search by cuisine or type
sqlite3 ~/places/places.sqlite "SELECT id, name, location, price_range, tags FROM places WHERE cuisine_or_type LIKE '%italian%';"

# Search by tag
sqlite3 ~/places/places.sqlite "SELECT id, name, category, location FROM places WHERE tags LIKE '%date-night%';"

# Search by location
sqlite3 ~/places/places.sqlite "SELECT id, name, category, cuisine_or_type FROM places WHERE location LIKE '%seattle%';"

# Places we haven't visited
sqlite3 ~/places/places.sqlite "SELECT id, name, category, location, notes FROM places WHERE visited_at IS NULL ORDER BY created_at DESC;"

# Places we've visited and rated
sqlite3 ~/places/places.sqlite "SELECT name, category, location, rating, visited_at FROM places WHERE visited_at IS NOT NULL ORDER BY rating DESC;"

# Get full place details by ID
sqlite3 ~/places/places.sqlite "SELECT * FROM places WHERE id = 5;"

Marking a place as visited

sqlite3 ~/places/places.sqlite "UPDATE places SET visited_at = datetime('now') WHERE id = ID;"

Rating a place after visiting

sqlite3 ~/places/places.sqlite "UPDATE places SET rating = 4, visited_at = COALESCE(visited_at, datetime('now')), notes = COALESCE(notes || ' | ', '') || 'Great pasta, noisy on weekends.' WHERE id = ID;"

Recommendation queries

# Date night restaurants we haven't tried
sqlite3 ~/places/places.sqlite "SELECT name, location, cuisine_or_type, price_range, notes FROM places WHERE tags LIKE '%date-night%' AND category = 'restaurant' AND visited_at IS NULL;"

# Highly rated places to revisit
sqlite3 ~/places/places.sqlite "SELECT name, category, location, rating, notes FROM places WHERE rating >= 4 ORDER BY rating DESC;"

# Weekend activity ideas
sqlite3 ~/places/places.sqlite "SELECT name, cuisine_or_type, location, notes FROM places WHERE category IN ('activity', 'park', 'attraction') AND visited_at IS NULL;"

Schema reference

places (
  id INTEGER PRIMARY KEY,
  name TEXT NOT NULL,
  category TEXT,              -- restaurant, cafe, bar, activity, park, attraction, shop
  source_url TEXT,
  source_platform TEXT,       -- instagram, google, yelp, tiktok, website, manual
  location TEXT,              -- neighborhood/city or address
  cuisine_or_type TEXT,       -- cuisine for restaurants, type for activities
  tags TEXT,                  -- JSON array
  price_range TEXT,           -- $, $$, $$$, $$$$
  notes TEXT,
  saved_by TEXT,
  visited_at DATETIME,
  rating INTEGER,             -- 1-5 after visiting
  created_at DATETIME
)

Trigger phrases

"save this place"
"remember this restaurant"
"add this to my places list"
"where should we go for dinner?"
"what places has wife saved?"
"show me date night spots"
"we went to [place] last night" (mark visited)
"rate [place] 4 stars"
"places we haven't been to"
Any restaurant/place URL shared with a request to save