All Docs
FeaturesSidekickUpdated March 11, 2026

Destructive Action Confirmation Gate

Destructive Action Confirmation Gate

Sidekick agents are designed to work autonomously on your behalf, but certain actions carry risks that cannot be undone. The Destructive Action Confirmation Gate is a permanent safety layer that ensures a human is always in the loop for these actions — regardless of how your agent's autonomy level is configured.

What It Is

The confirmation gate is a hard stop built into the agent execution pipeline. When the agent plans an action that falls into a protected category, it pauses execution and surfaces an approval request before proceeding. This behaviour is enforced at the platform level and cannot be overridden by autonomy settings.

Always-Confirmed Action Categories

🗑 Data Deletion

Any action that would permanently remove data — including deleting emails, files, calendar events, database records, chat messages, or repository content — requires confirmation.

📨 Sending to New Contacts

If the agent intends to send an email, message, or any outbound communication to a recipient it has not previously contacted on your behalf, the action is held for approval. This prevents unintended outreach to unknown parties.

💳 Financial Transactions

Actions that initiate, authorise, schedule, or modify any payment, money transfer, invoice, or subscription are always confirmed. This includes actions taken via connected services such as payment processors or banking integrations.

🔐 Access Revocation

Revoking OAuth tokens, disconnecting integrations, removing granted permissions, or signing out of connected accounts will always trigger a confirmation prompt before the agent proceeds.

How It Works

  1. The agent identifies an intended action as part of a task
  2. If the action matches a protected category, execution is paused
  3. An approval prompt appears in your Sidekick dashboard
  4. You approve or reject the action
  5. The agent resumes (or discards) the action and continues the task

All confirmation requests and your responses are recorded in the Activity Feed for audit purposes.

Autonomy Level Interaction

The confirmation gate operates independently of your autonomy level setting. Even an agent configured for fully-autonomous operation will pause and request approval for gated actions. This is intentional — the gate exists precisely because these actions are high-risk regardless of how much you trust the agent in other contexts.

Autonomy LevelStandard ActionsGated Actions
ManualRequires approvalRequires approval
Semi-autonomousExecutes automaticallyRequires approval
Fully autonomousExecutes automaticallyRequires approval

Frequently Asked Questions

Can I disable the confirmation gate? No. The gate is enforced at the platform level and cannot be disabled through autonomy settings or any other configuration option.

What happens if I don't respond to a confirmation request? The action remains pending. The agent will not proceed until you explicitly approve or reject it. Pending confirmations are visible in your dashboard.

Does the gate apply to all integrations? Yes. The gate applies across all connected services and integrations — it is enforced at the agent execution layer before any integration-specific logic runs.

Where can I see past confirmation requests? All confirmation events — including the action requested, your response, and the timestamp — are logged in the Activity Feed on your dashboard.