> ## Documentation Index
> Fetch the complete documentation index at: https://docs.epode.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Feedback modes

> Choose when customer agents submit feedback and when they must ask permission.

Configure the product's mode in **Collection policy**, then deploy the matching
`AGENT_FEEDBACK_MODE` value so the emitted instruction stays in sync.

| Dashboard label | Environment value | Agent behavior                                                                        |
| --------------- | ----------------- | ------------------------------------------------------------------------------------- |
| Never ask       | `never_ask`       | Submit the agent's own bounded assessment without interrupting the user               |
| Ask once        | `ask_once`        | Ask once for this product in this agent runtime; remember approval or refusal locally |
| Ask every time  | `ask_always`      | Ask before each report                                                                |
| Off             | `off`             | Emit no feedback contract and reject reports                                          |

## Ask once

The response includes an opaque `consentScope`. The agent runtime—not Epode—stores approval or
refusal under that scope. A later approved report uses `approvalSource: "stored_grant"`.

Rotating a product key produces a new scope, so an agent may ask again. The scope is not a user or
agent identity.

## Ask every time

Every report requires fresh approval and `approvalSource: "granted_now"`. Stored consent is ignored.

## Permission copy

Use the concrete question carried by the protocol:

> May I send the product provider a short feedback report about how this product worked? Your prompt and task content will not be included.

<Warning>
  Never infer approval from silence. Diagnostics validate ask-mode contracts but do not submit a report,
  because a diagnostic cannot impersonate the user.
</Warning>