AI agent playbooks
Read-only account health check
Gather account status within the connection's scope and report issues that need a person without changing configuration.
Goal
You check the accounts the connection may view and produce an evidence-based report. This playbook is read-only: do not sign in, change proxies, enable accounts or test by sending messages.
Permissions needed
accounts.readto view accounts.stats.readif you need health or limit statistics.audit.readif you need to cross-check alerts and logs.
Infrastructure read permissions do not automatically include account read permissions. Only use operations the current connection is actually allowed to call.
Ask the person before
Define the scope: which accounts to check, whether the task is a status report or investigation of a specific incident, and who receives the report.
If the person asks you to “fix it too,” separate the fix from this check. A request to check is not consent to sign in again, change machines or change proxies.
Check first
Call whoami to confirm the workspace, connection and granted accounts. Use tools_search to find the required operations and tools_describe to read their schemas before calling them.
Returned lists only reflect what you can see. Do not conclude that the workspace has no other accounts if the connection has a restricted account scope.
Steps
- Call
accounts_list, corresponding to RESTaccounts.list. - Follow pagination until you cover the agreed scope.
- Call
accounts_getfor each account to check, corresponding toaccounts.get. - If permitted, read
stats.healthwithtools_run. - If investigation is needed, read
alerts.listwithtools_run. - Read
audit.listwhen you need to cross-check recent changes. - Summarize each account: observed data, anomalies and decisions that need a person.
Do not turn a machine or connection status into a definite conclusion about a Telegram session. Report account, session, proxy and machine aspects separately when the results distinguish them.
Check after
The report must state when you checked, what scope you covered and which parts you lacked permission to read. Distinguish “no errors were returned” from “fully checked.”
Do not include keys, login sessions or secret proxy information in the report. You may use account names in an authorized environment, but externally shared documents need fake or redacted data.
On errors
If an account is no longer accessible, record the permission limit instead of trying another identity. If a read fails, do not infer that the account is broken.
When rate-limited, wait as directed by the response. Do not open additional connections to keep reading faster.
When to stop
Stop when you have checked the full scope, the requester cancels, the connection loses permission or there is not enough data to reach a safe conclusion. The final result is a report, not a list of changes you made on your own.
See also safety states and account management.