Health Accountability Sub-Agent: Fitbit + InfluxDB daily check-in with escalating nudges

[BenSpex]

Overview

Build a daily accountability sub-agent that queries Ben's Fitbit data from InfluxDB, cross-references manual check-ins for pillars that can't be auto-detected, scores each day, and delivers increasingly pointed feedback the longer streaks break.

Stack: InfluxDB at http://influxdb:8086 · Grafana at http://192.168.77.125:3000 · bens_agents chat integration

---

The Five Pillars

Pillar	Data Source	Detection Method
Sleep	Fitbit via InfluxDB	Auto — sleep score, minutes asleep, sleep stages, HRV
CrossFit	Fitbit via InfluxDB	Auto — Active Zone Minutes, exercise type, duration
Ice Bath	Manual	Morning check-in prompt (no sensor proxy)
Meditation	Manual + HRV proxy	Check-in + flag if HRV trending down (stress signal)
Food	Manual + meal_prep agent	Check-in + cross-reference meal_prep plan

---

InfluxDB Data Layer

health/influx_client.py

Thin wrapper around influxdb-client-python:

from influxdb_client import InfluxDBClient

INFLUX_URL = "http://influxdb:8086"
INFLUX_TOKEN = os.getenv("INFLUX_TOKEN")  # store in .env
INFLUX_ORG = "bens_health"
INFLUX_BUCKET = "fitbit"  # confirm bucket name on first run

client = InfluxDBClient(url=INFLUX_URL, token=INFLUX_TOKEN, org=INFLUX_ORG)

Key Flux Queries

Sleep score for last night:

from(bucket: "fitbit")
  |> range(start: -24h)
  |> filter(fn: (r) => r._measurement == "sleep" and r._field == "score")
  |> last()

Active Zone Minutes (CrossFit proxy):

from(bucket: "fitbit")
  |> range(start: -24h)
  |> filter(fn: (r) => r._measurement == "activity" and r._field == "active_zone_minutes")
  |> sum()

Resting HR + HRV (stress / recovery signal):

from(bucket: "fitbit")
  |> range(start: -7d)
  |> filter(fn: (r) => r._measurement == "heart_rate"
      and (r._field == "resting_heart_rate" or r._field == "hrv"))
  |> mean()

First-run note: Run list_measurements() against the InfluxDB bucket to confirm actual measurement/field names from the Grafana data source before hardcoding.

---

Agent Architecture

health/accountability_agent.py

Morning run (7:00 AM)

1. Pull last night's Fitbit data — sleep score, HRV, resting HR
2. Auto-assess sleep pillar
3. Send morning check-in prompt to Ben via bens_agents chat:
   • "Ice bath today? (y/n)"
   • "CrossFit on the schedule?" (auto-confirms later if AZM > 30 detected)
   • "How's food looking today?"
4. Log responses to InfluxDB (accountability measurement)

Evening run (8:30 PM)

1. Pull day's activity data — AZM, steps, exercise sessions
2. Auto-confirm CrossFit pillar
3. Prompt for meditation and food close-out
4. Compute daily score (0–5 pillars met)
5. Update rolling compliance averages (7-day, 30-day)
6. Deliver verdict (see escalation model)

---

Escalation Model

Escalation driven by rolling 7-day compliance rate per pillar, tracked independently.

def get_tone(compliance_7d: float) -> str:
    if compliance_7d >= 0.85:   return "encouraging"
    elif compliance_7d >= 0.70: return "direct"
    elif compliance_7d >= 0.50: return "pointed"
    else:                        return "ruthless"

Tone Examples — Sleep pillar

Tone	Example message
Encouraging	"6h42m last night — not your best but you're tracking well overall."
Direct	"Three nights under 7 hours this week. That's not a blip, it's a pattern."
Pointed	"5 of 7 nights under 7 hours. Your HRV is showing it. What's actually going on?"
Ruthless	"18 days. Sleep compliance is 38%. You're not tired because of work — you're tired because you keep doing this."

Each pillar escalates independently. Crush sleep but dodge ice baths and the ice bath message gets brutal while sleep stays warm.

---

InfluxDB Write-back (Accountability Scores)

measurement: accountability
tags:
  pillar: sleep | crossfit | ice_bath | meditation | food
fields:
  met: 0 or 1
  score_raw: float        # e.g. Fitbit sleep score 82, AZM 45
  tone_level: 0-3
  streak_missed: int
  compliance_7d: float
  compliance_30d: float

Grafana panel: compliance heatmap + streak tracker per pillar.

---

Integration with Jenna (bens_agents)

• New tool: check_health_accountability() added to chat_tools.py
• Returns: last night's sleep score, yesterday's pillar completion, outstanding check-ins
• Jenna surfaces it in morning briefings naturally:

  "Sleep score was 74 — HRV dipped, probably the late call. Ice bath still unconfirmed."

• If Ben's on a multi-day miss streak on any pillar, Jenna flags it unprompted

---

Implementation Checklist

• Confirm InfluxDB bucket name + actual Fitbit measurement/field names (list_measurements query)
• Set up INFLUX_TOKEN in .env / NAS secrets
• Build health/influx_client.py with typed query helpers per pillar
• Build health/accountability_agent.py with morning + evening flows
• Implement escalation model + tone engine
• Write accountability scores back to InfluxDB
• Add check_health_accountability() tool to chat_tools.py
• Create Grafana compliance panel (heatmap + streak tracker)
• Schedule via bens_agents cron (7:00 AM + 8:30 PM daily)
• Close duplicate issues on bens_agents repo (#12 and #13)

---

Open Questions

• What's the InfluxDB auth token? (check NAS .env or Grafana data source config)
• Is the bucket named fitbit or something else?
• For ice bath + meditation check-ins: chat prompt, Telegram, or push notification?

Reference video: https://www.youtube.com/watch?v=VGY_XMXMSZc