Set rules so your AI agent uses official connections, runs at a sensible pace and stops at any security check, so your own accounts do not get restricted.
Who it is for
Owners whose agent logs into social, ad or marketplace accounts to read insights or post, and who cannot afford a locked account.
Tools
Your agent's standing instructions
A list of the accounts it touches
The platform's official app, API or export, where one exists
Steps
List every account the agent logs into and what it does there.
Check each platform's terms. Meta's say you may not access or collect data from its products by automated means without prior permission.
Switch to an official route where one exists: a connector, the platform's own export, or its API.
Add the rules from the prompt below to the agent's instructions.
Turn on two-factor sign-in and keep a recovery method you control, in case of a lock.
If an account gets flagged, stop the agent on that account before appealing.
The prompt
Add these rules to your standing instructions for any website account:
1. Use only the connections I list: [OFFICIAL APPS, APIS, EXPORTS].
2. Do not click through a site page by page to collect data.
3. Run each account task at most [N] times a day. Never loop a task that fails; stop after 2 tries and tell me.
4. At any captcha, security check, new-device prompt or warning: stop immediately and tell me. Never try to get past it.
5. Never change account settings, names or security options.
Tips
Exports and official reports usually give the same insights without the risk.
A loop on a failing task looks most like a bot; cap retries.
Keep the agent off the account you can least afford to lose.