Safety

Webhook content is what a webhook’s sender wrote: a request’s body, headers, path and query. Read this page before you let an agent read it and act on it.

How it works

An agent reads webhook content as Hookspot stored it, Authorization and Cookie headers included. Only credentials that verification removed are missing. Your app’s responses can echo the request, so treat them the same way.

Anyone who knows a source’s webhook URL can send it text. That text can be written to look like instructions to your agent. Hookspot returns it only as data, but an agent can still be misled by it.

Everything the agent reads goes to its provider, such as Anthropic or OpenAI, as any other text in the chat does.

Debug on a Read-only app

Approve an agent as Read-only when it only needs to find out why a webhook failed. A misled Read-only agent can read, but it cannot change your projects or send webhooks.

Connect a second app without Read-only for setup work, and revoke it when you are done.

Check changes before you approve them

Most agents ask before they run a tool that makes changes. Read the tool name and its arguments before you approve, especially after the agent read webhook content.

These tools send a webhook to your app again, which can make your app act twice, such as charge a card or send an email:

• retry_request
• replay_request
• retry_delivery
• retry_skipped_deliveries
• send_test_request

What agents never get

A source’s verification secret is never returned. An agent sees only whether one is set.

Some changes stay in the dashboard, as Dashboard links lists.