Inbox Agent: commitments, todos, and stale threads

Closed-loop email hygiene:

• Open commitments surfaced as a first-class surface—so promises do not live only as mental load.
• Extract todos flows that turn language like “send the proposal by Tuesday” into trackable follow-ups instead of a label only.
• Follow-up action to surface stale threads that still need a response—pair with your own cadence (weekly review, end-of-day).
• Refresh intel on a selected thread when new replies arrived and you want summary + commitment extraction + contact signals updated together.

Details and limitations are documented under Inbox Agent.

• Commitments are actionable in principle—accepted items can move toward real follow-up tasks instead of staying as a tag.
• Local persistence keeps thread state visible across restarts; you are not re-deriving context from scratch each session.
• Review gates still apply: external actions remain tied to provider rules and your confirmation habits.

1. Start each week from the open commitments card—what is overdue, what is ambiguous?
2. For dense threads, Refresh intel after new messages rather than trusting an old summary.
3. When Extract todos proposes items, merge duplicates and assign owners in your real task system.
4. Run Follow-up before you declare inbox zero—silence is not resolution.
5. Close the loop: mark done in both mail and your task tracker to avoid two sources of truth drifting.

• A task system you trust (even a simple list) for items that leave the inbox.
• Naming conventions so “todo from email” is searchable later.

1. Inventory recurring commitment patterns (sales, finance, HR).
2. Template responses for the top five—Prep thread gets faster.
3. Measure time from commitment to closure; tune Follow-up frequency.
4. Audit monthly: false positives in commitments (over-extraction).
5. Celebrate boring weeks—zero drama means the system is working.

• “List commitments in this thread with owner and deadline if stated.”
• “Which threads are stale but still socially important?”
• “What one follow-up would unblock the other party?”

• Legal or HR threads have explicit human review before any automated todo export.
• SLA expectations match what Follow-up can surface—do not promise sub-minute SLA from email alone.

• Over-extraction of todos—noise in the task system.
• Duplicate tracking: same promise in email, Slack, and JIRA—pick one system of record.
• Ignoring threads that are emotionally hard but not “stale” by time.
• Skipping Refresh intel after a heated reply chain—summaries go stale fast.