Analyze Travel Url

Analyze Travel URL

Use this skill when the input is a travel-related URL and the goal is to turn it into structured records for this project.

This project already has:

• functions/api/analyze-url.js for URL analysis
• functions/api/confirm-analysis.js for confirmed Notion write-back

Read these supporting files before deciding the final output:

• notion-field-mapping.md
• output-schema.md
• extraction-checklist.md

Use examples.md when the source contains multiple places, mixed spot/event content, or weak evidence.

Input

Primary input: $ARGUMENTS

If $ARGUMENTS is empty, use the URL provided in the current user request.

Goal

Produce a structured analysis result that: 1. aligns with the project's current analysis model 2. expands multiple places into separate items 3. distinguishes source-level information from item-level information 4. recommends whether the result should go to:

• Notion Sources
• Notion Spots
• Notion Events
• quick-access notes
• handoff notes

Required working order

Step 1: Identify source type

Classify the URL as one of:

• short-form video
• article / blog
• event page
• map / directory page
• weak-information page

Step 2: Collect evidence

Prefer evidence in this order: 1. structured page metadata 2. captions / subtitles 3. audio transcript 4. visible text in frames 5. title / description / user notes

If tool-assisted audio or visual extraction is unavailable in the current environment, explicitly mark the extraction as partial and continue with metadata plus notes.

Never pretend audio or visual evidence exists when it does not.

Step 3: Extract source-level facts

Extract:

• source title
• source platform
• probable city slug
• probable area
• overall content kind
• source credibility
• concise summary
• whether review is required

Step 4: Extract candidate items

Identify candidate places or events mentioned in the URL content.

For each candidate item, extract:

• name
• item kind
• category
• description
• tags
• city slug
• area
• best time or dates
• estimated stay or event duration clues
• map / official URL if available
• evidence
• source credibility
• item confidence
• review reason if uncertain

Step 5: Expand multiple locations

If a single source contains multiple places, create one item per place.

Rules:

• do not merge different places into one record
• do not collapse multiple restaurants into a single “food list” item
• if the source mixes places and events, keep them as separate items
• if the source mentions an itinerary order, preserve that order in reasoning notes

Step 6: Decide write targets

Always prepare:

• one source-level object
• zero or more item-level objects

Recommend write targets as follows:

• notion_source: always, unless the URL is invalid or unusable
• notion_spot: when an item is a stable place or store
• notion_event: when an item is date-bound or clearly an event
• quick_access: only when the source reveals a reusable rule, repeated extraction pattern, or structural learning worth preserving
• handoff: when ambiguity remains, manual review is needed, or follow-up enrichment is required

Step 7: Recommend schema updates only when necessary

If the current project schema cannot faithfully represent the source, recommend the smallest useful schema change.

Prefer additive fields over renaming existing fields.

Output contract

Return five sections in this order:

1. extraction_summary 2. source_record 3. items 4. write_recommendations 5. schema_recommendations

Hard rules

• Split multiple locations into separate items.
• Do not invent coordinates, dates, or official URLs.
• If evidence is weak, set review flags rather than fabricating certainty.
• Use current project field names unless explicitly proposing schema upgrades.
• If the source is too weak to support item creation, return source_only.

Examples

Example 1: one reel, three cafes

Input

Instagram reel showing:

• Cafe A interior
• Cafe B dessert
• Cafe C exterior sign

Caption mentions "京都咖啡三選"

Expected handling

• classify source as short_video
• classify overall content_kind as spot
• create one source record
• create three separate spot items
• each cafe gets its own:
• name
• item_kind=spot
• category=咖啡
• description
• tags
• area
• best_time
• stay_minutes
• evidence
• write_targets=["notion_spot"]

Wrong handling

• one combined item called "京都咖啡三選"

---

Example 2: one short video mixing a market and a seasonal event

Input

Video mentions:

• local night market
• weekend flower festival
• event date shown in video text

Expected handling

• one source record
• one spot item for the market
• one event item for the flower festival
• market -> write_targets=["notion_spot"]
• festival -> write_targets=["notion_event"]
• if date text is partly unclear, set needs_review=true and keep the observed evidence

---

Example 3: weak source with only a title and no clear place

Input

A URL with title like "Tokyo hidden gems" No readable description No extractable place names No notes

Expected handling

• content_kind="source_only"
• no item creation
• source record only
• recommend handoff if user may need manual completion

---

Example 4: article listing five restaurants with districts

Input

Blog article with five restaurant headings and district names

Expected handling

• source type = article
• content kind = spot
• one source record
• five separate spot items
• item order should follow article order
• use district names as area
• if map links exist, retain them

Extraction Checklist

Use this checklist before finalizing the structured result.

A. Source validation

• Is the URL valid and reachable?
• Is the source actually travel-related?
• Is there enough information to support item extraction?
• If not, should it remain source_only?

B. Evidence collection

• Page title collected
• Meta description collected
• OG title / description collected
• User notes considered
• Captions / subtitles available
• Audio transcript available
• Visible text from frames available
• Dates / venue names / district names identified
• Unsupported evidence explicitly marked as unavailable

C. Source-level analysis

• Source title normalized
• Platform identified
• Overall content kind chosen
• City slug inferred or left null
• Area inferred or left null
• Confidence set
• Review flag set
• Summary written

D. Item expansion

• All distinct places expanded into separate items
• Mixed place/event content separated
• Duplicates merged only when truly the same entity
• Each item has at least:
• name
• category
• description
• city_slug or null
• area or null
• evidence
• confidence
• review flag

E. Write target decision

• Source-level record marked for notion_source
• Stable locations marked for notion_spot
• Date-bound items marked for notion_event
• Reusable extraction pattern marked for quick_access
• Ambiguous / incomplete result marked for handoff

F. Guardrails

• No invented dates
• No invented coordinates
• No invented official URLs
• No merged multi-location record
• No false certainty from weak evidence

Notion Field Mapping

This project currently writes confirmed analysis results into separate Notion data sources.

1) Source-level record

Use for every analyzed URL.

Target

Notion Sources data source

Current field mapping

• Name <- short display title derived from city + summary or platform + kind
• SourceUrl <- normalized source URL
• Platform <- source platform
• SourceType <- "活動資訊" | "景點美食" | "手動整理"
• Status <- current workflow status
• Note <- user notes or summary
• CityHints <- inferred city slug(s)
• Published <- boolean

Skill-side source object fields

Use these field names in the analysis result:

• source_title
• source_platform
• content_kind
• city_slug
• area
• confidence
• needs_review
• summary
• source_credibility
• review_reason
• write_targets

2) Spot-level item

Use when the item is a stable place, venue, restaurant, cafe, shopping point, temple, museum, etc.

Target

Notion Spots data source

Current field mapping

• Name <- item name
• Area <- area or fallback city label
• BestTime <- best visit time
• Category <- category
• City <- city label
• CitySlug <- city slug
• Description <- short description
• Lat <- coordinate
• Lng <- coordinate
• MapUrl <- map link
• Notes <- reason / analyst note
• Published <- boolean
• StayMinutes <- stay estimate
• Tags <- joined tags
• Thumbnail <- emoji thumbnail

Skill-side spot item fields

• name
• item_kind = spot
• category
• description
• tags
• city_slug
• area
• best_time
• stay_minutes
• lat
• lng
• map_url
• thumbnail
• source_credibility
• item_confidence
• needs_review
• review_reason
• evidence
• reason

3) Event-level item

Use when the item is date-bound, ticketed, time-limited, or clearly an event.

Target

Notion Events data source

Current field mapping

• Name <- item name
• Area <- area
• Category <- category
• City <- city label
• CitySlug <- city slug
• Description <- short description
• StartTimeText <- start time
• EndTimeText <- end time
• StartsOn <- start date
• EndsOn <- end date
• Lat <- coordinate
• Lng <- coordinate
• MapUrl <- map link
• OfficialUrl <- official or source URL
• PriceNote <- price note
• Published <- boolean
• RecurringType <- default recurrence
• Status <- workflow status
• Tags <- joined tags
• TicketType <- ticket mode
• VenueName <- venue name

Skill-side event item fields

• name
• item_kind = event
• category
• description
• tags
• city_slug
• area
• starts_on
• ends_on
• start_time
• end_time
• price_note
• ticket_type
• venue_name
• lat
• lng
• map_url
• official_url
• thumbnail
• source_credibility
• item_confidence
• needs_review
• review_reason
• evidence
• reason

4) Current model gap to be aware of

The existing backend already supports:

• source-level contentKind
• item arrays
• spot/event-specific item properties
• Notion write-back by content kind

However, current analysis output should ideally be extended with:

• item_kind
• source_credibility
• item_confidence
• evidence
• review_reason

Recommend these only when needed. Do not force a schema rewrite if the current task can be completed safely without them.

Output Schema

Return the result in JSON-like structure using the following shape.

Top-level object

{
  "extraction_summary": {
    "source_type": "short_video | article | event_page | map_page | weak_information",
    "extraction_mode": "audio_visual | metadata_only | metadata_plus_notes | mixed",
    "source_credibility": "high | medium | low",
    "overall_confidence": 0.0,
    "needs_review": true,
    "review_reason": "string"
  },
  "source_record": {
    "source_title": "string",
    "source_platform": "Instagram | Threads | Facebook | YouTube | TikTok | Website",
    "content_kind": "spot | event | source_only",
    "city_slug": "string or null",
    "area": "string or null",
    "confidence": 0.0,
    "needs_review": true,
    "summary": "string",
    "source_credibility": "high | medium | low",
    "review_reason": "string",
    "write_targets": ["notion_source"]
  },
  "items": [],
  "write_recommendations": [],
  "schema_recommendations": []
}

{
  "name": "string",
  "item_kind": "spot | event | source_only",
  "category": "string",
  "description": "string",
  "tags": ["string"],
  "city_slug": "string or null",
  "area": "string or null",
  "best_time": "string or null",
  "stay_minutes": 0,
  "starts_on": "YYYY-MM-DD or null",
  "ends_on": "YYYY-MM-DD or null",
  "start_time": "HH:MM or empty",
  "end_time": "HH:MM or empty",
  "price_note": "string or null",
  "ticket_type": "string or null",
  "venue_name": "string or null",
  "lat": 0,
  "lng": 0,
  "map_url": "string or null",
  "official_url": "string or null",
  "thumbnail": "emoji or null",
  "source_credibility": "high | medium | low",
  "item_confidence": 0.0,
  "needs_review": true,
  "review_reason": "string",
  "evidence": [
    {
      "type": "audio | visual_text | caption | title | description | metadata | note",
      "value": "string"
    }
  ],
  "reason": "string",
  "write_targets": ["notion_spot"]
}

{
  "target": "notion_source | notion_spot | notion_event | quick_access | handoff",
  "scope": "source | item",
  "name": "source title or item name",
  "why": "string"
}

{
  "field": "string",
  "level": "source | item",
  "reason": "string",
  "priority": "high | medium | low"
}