I've been having trouble with my blood sugar spiking after lunch.
So I asked GitHub Copilot CLI to help me figure out what's going on
- and when it couldn't answer my question, we fixed the skill
together, live.
Using my Nightscout CGM data and a custom Copilot skill, I asked
"what was my worst lunch this week?" The AI found the problem
(yesterday, peaked at 294 mg/dL 😬) but needed to write a custom
Python script to do it. That's a sign the tool is missing
something!
So we added two new commands:
- day - drill into exactly what happened on a specific date
- worst - find your problem days ranked by peak glucose
We also removed hard-coded "meal presets" because everyone eats at
different times (and this needs to work globally). Now the AI just
asks "what hours do you eat lunch?" and uses those.
This is the magic of AI-assisted development: the AI uses the tool,
hits a limitation, and helps you fix it - all in the same
conversation.
🔗 Nightscout CGM Skill:
https://github.com/shanselman/nightscout-cgm-skill 🔗 Agent Skills
standard: https://github.com/agentskills/agentskills
#GitHubCopilot #Diabetes #Nightscout #T1D #AIAssistant #LiveCoding