Connection Checklists

Review the workflow before the agent touches real systems.

These guides are practical pre-connection reviews. They help teams decide what an agent can touch, what needs to be limited, and where human approval should stay in the loop.

Email

Before connecting agents to email

  • Separate read-only inbox access from send permissions.
  • Require approval before external replies, forwarding, or attachments.
  • Log every draft, sent message, recipient, and source thread.
  • Block access to sensitive folders unless explicitly needed.
CRM

Before connecting agents to CRM

  • Scope access by pipeline, role, location, or account type.
  • Limit create/update permissions until the workflow is reviewed.
  • Require approval before changing deal stage, owner, price, or status.
  • Track every field the agent reads or writes.
Calendar

Before connecting agents to calendars

  • Use availability lookup before granting booking authority.
  • Set booking windows, buffers, service areas, and cancellation rules.
  • Require approval for VIP, after-hours, or high-value appointments.
  • Log event creation, edits, invites, and cancellations.
Files

Before connecting agents to files

  • Start with a limited folder, not the full drive.
  • Separate read, write, delete, and share permissions.
  • Block secret keys, payroll, legal, medical, and customer-sensitive folders.
  • Record file names, paths, actions, and generated outputs.
Browser

Before connecting agents to browsers

  • Limit sites, forms, downloads, and login sessions.
  • Require approval before submitting forms or changing account settings.
  • Block purchases, password changes, and sensitive portals by default.
  • Capture screenshots, URLs, and action history for review.
GitHub / Code

Before connecting agents to code

  • Use branch and pull-request workflows before direct commit rights.
  • Keep production secrets and deployment credentials out of agent reach.
  • Require review for dependency, auth, payment, and data-handling changes.
  • Log prompts, diffs, commands, and tests run.
Payments

Before connecting agents to payments

  • Start with quote or invoice drafting, not autonomous payment execution.
  • Set spending limits, refund limits, and vendor restrictions.
  • Require human approval for charges, payouts, refunds, or bank changes.
  • Keep receipts, approval trails, and transaction IDs.
Baseline

Minimum controls

  • Scoped permissions.
  • Audit logs.
  • Human approval where needed.
  • Fallback path when the agent is unsure.

Trust is a deployment decision.

The goal is not to make broad promises. The goal is to decide whether a specific agent workflow is safe enough to connect under specific limits.

Review an AI Workflow