Pattern Radar
Two of the highest-leverage questions a busy knowledge worker can ask are “what should I write about?” and “what could I automate?” — and the answers are usually hiding in plain sight, in the work you already did this week.
Pattern Radar scans your recent Microsoft 365 signals — Inbox and Sent mail, active chats and channels, and calendar subjects — and surfaces recurring patterns worth turning into leverage:
- Blog candidates — things you keep explaining to different people.
- Automation candidates — multi-step tasks you keep doing by hand.
It’s read-only and privacy-aware: it never sends anything and never names customers or people in its output.
Prefer this to run on a schedule? There’s a companion Pattern Radar (Scheduled) automation that posts the same digest to Teams twice a week.
Before you start
- A connected Microsoft 365 account — the skill reads mail subjects, chat/
channel messages, and calendar subjects via the agent’s
m365_*-style tools. - Tool names vary by host; the skill describes capabilities, not one vendor’s exact tool names.
How to use it
Ask, e.g.:
- “/pattern-radar”
- “What should I write about this week?”
- “Find things I could automate from my recent work.”
It looks back over the last 7 days, clusters your signals, and returns 0–3 candidates. Each candidate names the theme, the anonymized evidence, whether it’s a blog or automation opportunity, and a one-line rationale (for automations, the trigger and the action).
Good to know
- Ruthless threshold. A pattern only counts if it shows up ≥3 times across ≥2 distinct people/chats/threads/events. One-offs are treated as noise, so a quiet week honestly returns “No new patterns.”
- Biased toward automation. Blog and automation candidates get equal weight, but when an automation candidate qualifies it’s preferred — offloading repetitive work is usually the bigger win.
- Sent mail is gold. It pays special attention to what you send (forwards, intros, status pings, “looping in X” handoffs) — that’s where repetitive manual work hides.
- Privacy first. Read-only, always anonymized (“3 different customers”, never the actual names), and it skips anything you’ve marked private or excluded.