Creation state
Forced paused.
Every new campaign, ad group, ad set, and ad is created inactive. The rule lives in the execution layer, not in a reminder prompt.
Safety and control
Useful power. Hard boundaries.
A chat instruction is not a security policy. MoonSignal enforces the important rules inside the private engine—outside the model’s discretion.
OFFDefault creation state
0Delete endpoints
100%Approval visibility
Destructive access
Deletion is unavailable.
Delete tools are absent. MoonSignal can prepare, pause, and improve without erasing account objects or history.
Financial control
Limits are enforced.
Set the maximum daily or monthly spend. Requests beyond the limit are rejected or returned for explicit review.
Budget changes
Large jumps trigger warnings.
MoonSignal encourages measured increases and flags disruptive changes that may reset platform learning.
Approval
You authorize execution.
Every proposal states what will change, why, and the current status. Launch remains an owner decision.
Traceability
Actions leave a record.
Prepared and executed actions are tied to the client workspace so support and the owner can see what happened.
Private client environmentEach client’s connection and business configuration are isolated from other client workspaces.
Secrets stay out of chatAccount credentials and tokens are stored as encrypted service secrets, not pasted into conversations or returned in tool output.
Revocable accessDisconnect the integration or revoke platform permissions when the relationship ends. Access should not become permanent.
Minimum necessary permissionsRead and action access are scoped to what the connected workflow needs. Destructive capability is intentionally excluded.
Approval-aware executionDrafting and analysis can move quickly. Launch, spending, and meaningful budget changes remain inside the allowed approval policy.
What Claude sees
Enough context to help. Not the keys to everything.
The conversation receives the data and actions needed for the task. Credentials, internal secrets, and another client’s information are not part of the response.
MoonSignal’s safety promise should describe the controls that are actually deployed for the client. Any optional automation mode must be explicitly enabled and documented before use.
Fast work should still feel safe to approve.
We configure the spending limits and approval rules during installation.
Talk through your guardrails →