Post-Experience Rater

A 15-second micro check-in after an experience. The point is speed and low friction -- this should feel like texting a friend "how was it?", not filling out a form. Each rating becomes a Fulcra annotation that, over time, builds the user's taste graph (what a future restaurant recommender and other skills draw on).

Prerequisites

• Fulcra CLI authenticated (uv tool run fulcra-api auth login).
• Run with uv run --python 3.12.

The conversation (keep it fast)

1. "How was [place/experience]?"
2. If it's a restaurant: "Food, vibe, service -- quick 1-5 each?" (For a hotel, use room/location/service; for an event, just an overall score is fine. Use whatever 1-5 keys fit.)
3. "Would you go back?"
4. "Anything worth remembering?" -- one free-text line ("too loud for a date night", "the pasta was incredible, bar seats are better").

Infer tags (cuisine, neighborhood, occasion) and companions from what they say; don't interrogate. Two or three quick exchanges is the whole thing.

Save

Assemble a small JSON file and run save. --dry-run first if you want to show the user what's being stored; for this skill it's fine to save directly once you've reflected their answer back, since it's low-stakes and fast.

uv run --python 3.12 ~/.claude/skills/post-experience-rater/scripts/post_experience.py save --payload /tmp/rating.json

Payload schema

{
  "experience_type": "restaurant",
  "name": "Carbone",
  "scores": {"food": 5, "vibe": 4, "service": 3},
  "would_return": true,
  "notes": "Too loud for a date night but the pasta was incredible. Bar seats are better.",
  "companions": ["Jordan"],
  "tags": ["italian", "west village", "special occasion"]
}

Only name is strictly required. scores accepts any 1-5 keys appropriate to the experience type. Fill what the user gave you; don't invent scores they didn't state.

What save does

Writes one "Post-Experience Rating" moment annotation (created once, reused after) carrying the full JSON, and reports verified_matches. Only report success when ok is true / verified_matches >= 1 -- not from a bare HTTP code.

After saving

Reflect it back in one warm line ("Got it -- Carbone, 5/4/3, you'd go back, too loud for dates. Logged."). That's it. If they mentioned a noteworthy new person, you can offer to add them via the relationship-crm/event-debrief skills -- but don't push.

Privacy

Ratings and companions are personal. Don't surface them in a public/group context. Never print API tokens.

Where this fits

Part of the Fulcra personal concierge family (shared lib in ~/.fulcra-concierge/lib). These ratings are the raw material for a future restaurant-recommender and feed the preference signal that meeting-cadence-optimizer and evening-debrief complement on the time/energy side.