1. 1

    Start with one outcome

    Pick a single thing the Skill should accomplish (e.g., scheduling follow-ups).

  2. 2

    Choose the minimum tools

    Start with calendar and/or contacts. Add other tools only after behavior is stable.

  3. 3

    Define boundaries

    Decide what the Skill can read, what it can write, and when it must ask for approval.

Common safety rules

  • Require approval for writes (creating events, editing contacts) until you trust the behavior
  • Treat personal data as sensitive by default
  • Add a "handoff to a human" rule for ambiguous requests
  • Log and review edge cases during early rollout

Browse Skills