Preview mode for destructive actions

Any writable endpoint supports a dry run. Add preview: true to the request body and Crispy tells you exactly what would happen without doing it.

What the preview returns

• description: a plain-language summary of the action and its cost
• budget: your API calls used, limit, and remaining
• safety: for LinkedIn actions, the daily safety category (e.g. invitations or messaging) with today's used, limit, and remaining counts
• severity and irreversible: a risk classification. Sends, deletes, and campaign launches are high severity because they cannot be undone or start real outward-facing activity. Reversible config changes are low.
• confirm_token: a signed token representing exactly this action

Executing the previewed action

Submit the same request again with the confirm_token included. The token is valid for 5 minutes and is bound to the tool, the exact arguments, and your account. Change any argument and it is rejected with 'Args do not match previewed action'. Wait too long and you get 'Token expired (5-minute window passed)'. Tampered tokens fail signature verification outright.

That makes preview-confirm safe to build into agent workflows: the thing that executes is cryptographically guaranteed to be the thing that was previewed, and a stale approval cannot fire an hour later.

When to use it

Wire preview mode into anything where a human approves the action: Slack approval flows, internal dashboards, agent guardrails. For high-severity actions like sending invitations, deleting campaigns, or launching outreach, previewing first costs one extra request and removes an entire class of surprises.