Acira
Docs

Secure Puppeteer Agents

Puppeteer agents can take sensitive browser actions. Acira adds a policy layer before those actions execute.

Use this page when adding action mediation to Puppeteer-based agents.

Direct answers

Action mediation

Route navigation, input, clicks, file operations, and script evaluation through a policy decision before passing them to the browser.

Secret safety

Keep credentials outside the model context and substitute them only at the point of use when policy allows the action.

Approvals

Require human approval for purchases, account changes, admin actions, and sensitive data movement.

FAQ

What is the main risk in Puppeteer agents?

The main risk is that a model-driven workflow can execute browser actions that exceed the user's intent or organizational policy.