Personal Concierge

Personal Concierge (Orchestration)

The front door to the concierge. It stitches the individual skills into a coherent daily rhythm so the user gets one smooth experience instead of having to know which skill does what. Each sub-skill still works on its own; this wraps them.

First: check what's available

uv run --python 3.12 ~/.claude/skills/concierge-wrapper/scripts/status.py

Reports which concierge skills are installed, whether Fulcra auth works, and whether Attio is configured. Use this for graceful degradation: run what's present, and if something's missing, do the rest and tell the user plainly what was skipped and why ("skipped meeting prep -- Attio isn't connected"). Never abort the whole sequence because one piece is unavailable.

Morning sequence -- "start my day"

Run as one flowing interaction, not a menu:

1. subjective-checkin -- the warm 2-3 question check-in (how they slept/feel). This writes the Morning Check-In annotation.
2. morning-briefing -- immediately after, run collect.py and deliver the briefing, calibrated to BOTH last night's sleep and what they just told you. Because the check-in ran first, the briefing can open with how they feel, not just the data.

Deliver it as a single continuous experience: the check-in naturally leads into "okay, here's your day." Don't make them ask twice.

Evening sequence -- "wrap up my day"

1. evening-debrief -- the end-of-day reflection (rating, cadence, action items, tomorrow preview). Writes the Evening Debrief annotation.
2. If it's Sunday or Monday (or the user asks), also run meeting-cadence-optimizer (analyze) and fold the insight into the wrap-up: "for the week ahead -- your sweet spot's been ~4 meetings/day."

Day-of routing (contextual)

When the user surfaces one of these mid-day, route to the matching skill (no need to run a whole sequence):

Weekly rhythm

• Sunday eve / Monday morning: meeting-cadence-optimizer, surfaced in the next morning briefing.
• Encourage the daily loop (check-in + debrief) -- the cadence analysis and the taste/relationship graphs only get sharp with consistent use. It's fine to gently note when there isn't enough history yet.

Graceful degradation -- the rules

• One sub-skill failing or missing never blocks the others. Skip it, name it, continue.
• Each sub-skill is independently usable via its own triggers; the wrapper is a convenience, not a dependency.
• Always be honest about what actually ran and what was stored vs. skipped -- don't imply a briefing happened if the data wasn't there.

Privacy

This sequence touches the most intimate data the concierge has -- feelings, sleep, calendar, relationships. Never expose specifics in a public/group context. Never print API tokens or secrets.

Where this fits

The orchestration layer over the whole Fulcra concierge family (shared lib in ~/.fulcra-concierge/lib): subjective-checkin, morning-briefing, post-experience-rater, evening-debrief, meeting-cadence-optimizer, event-debrief, relationship-crm. (A restaurant-recommender is a planned addition.)