> ## Documentation Index
> Fetch the complete documentation index at: https://hc.saas-manager.biz.nuro.jp/llms.txt
> Use this file to discover all available pages before exploring further.

# Review and apply AI improvement proposals

> How to review, edit, apply, and regenerate AI-generated Playbook improvement proposals after session completion

After each session closes, AI Agent automatically analyzes the conversation and generates improvement proposals for Playbooks. Operators review these proposals and choose to apply, reject, or regenerate them, continuously improving support quality.

<Info>
  Improvement proposals are generated automatically, but **applying them to Playbooks always requires an explicit operator action**. Proposals are never applied automatically.
</Info>

## How proposals are generated

When a session closes, AI Agent analyzes the conversation history in the background. If improvements to existing Playbooks or new Playbooks are detected, proposals are generated.

| Proposal type         | Description                                            |
| --------------------- | ------------------------------------------------------ |
| Playbook modification | A suggested change to an existing Playbook             |
| New Playbook          | A suggested new Playbook for a pattern not yet covered |

<Info>
  If existing Playbooks already sufficiently cover the session's pattern, no proposal is generated. This is not an error — it means the current Playbooks are working well.
</Info>

## Find and review proposals

Improvement proposals aren't collected on a single, dedicated screen. Check either of the following instead.

<Steps>
  <Step title="From an AI Playbook's detail view">
    Under **Knowledge > AI Playbook**, open the Playbook you want to check and select the **Proposals** tab. If there are pending proposals, a count badge appears on the tab.
  </Step>

  <Step title="From a ticket (session) detail view">
    To review the proposal generated from a specific session, open that ticket and select the **Proposals** tab.
  </Step>

  <Step title="Review the proposal list and details">
    Each **Proposals** tab lists pending proposals as cards. Each entry shows:

    | Field                 | Description                                         |
    | --------------------- | --------------------------------------------------- |
    | Proposal type         | Update Procedure or New Procedure                   |
    | Reason and background | Why this change is proposed                         |
    | Priority              | High / Medium / Low                                 |
    | Quality check result  | Pass / Needs Revision / Reject (automated AI check) |

    Click a proposal to view the diff between the current Playbook content and the proposed change, along with the proposal's reason and background. Review this information before taking action.
  </Step>
</Steps>

## Accept a proposal as-is

If the proposal looks appropriate, you can apply it directly to the Playbook.

<Steps>
  <Step title="Select a proposal and review the diff">
    Click a proposal from the list and review the diff and reason.
  </Step>

  <Step title="Accept">
    Click **Accept**.
  </Step>
</Steps>

When applied:

* The Playbook content is updated with the proposed content
* The updated Playbook becomes immediately active
* The proposal is removed from the list

## Edit and accept a proposal

If the proposal's direction is correct but the content needs adjustment, you can edit it before applying.

<Steps>
  <Step title="Select a proposal">
    Click a proposal from the list.
  </Step>

  <Step title="Open the editor">
    Switch from the diff view to edit mode. An editor pre-filled with the proposed draft content opens.
  </Step>

  <Step title="Adjust the content">
    Freely modify the proposed content in the editor.
  </Step>

  <Step title="Accept">
    Click **Accept** when editing is complete.
  </Step>
</Steps>

<Info>
  If you close the editor without applying, all edits are discarded. The proposal remains in the list unchanged.
</Info>

## Discard a proposal

If the proposal is not suitable, you can explicitly discard it.

1. Click the proposal from the list
2. Click **Discard**

Discarding removes the proposal from the list. The Playbook is not changed. If you want a revised version of the proposal, use **Regenerate** before discarding. Once a proposal is discarded, you can no longer regenerate it.

## Regenerate a proposal with feedback

If the proposal's direction is wrong, you can request a new proposal with feedback guidance.

<Steps>
  <Step title="Select a proposal">
    Click a proposal from the list.
  </Step>

  <Step title="Select Regenerate">
    Click **Regenerate**.
  </Step>

  <Step title="Enter a Feedback Prompt (optional)">
    Optionally enter a one-time text prompt to guide the regeneration. This prompt is not saved after the request is submitted.

    Feedback Prompt examples:

    * "Reference the VPN setup Playbook instead of the general troubleshooting Playbook."
    * "Add a VIP escalation check as the first step."
  </Step>

  <Step title="Confirm regeneration">
    Click **Regenerate**. The current proposal is removed from the list, and a new proposal is generated in the background.
  </Step>
</Steps>

## Leave feedback on a session via chat command

Separately from the proposal review screen, you can leave feedback on a completed session directly from chat. Use the `/feedback` command in Slack, Teams, or another connected chat tool. Each session accepts only one feedback comment. Once you submit it, you cannot submit a second comment for the same session. Submitted feedback appears on the session detail view in the admin screen.

Examples of session feedback:

* "For this type of inquiry, the AI should have referenced the VPN setup Playbook."
* "This session needed earlier escalation — the user was a VIP."
* "The resolution steps used here should become a standard Playbook for printer issues."

<Info>
  Submitting feedback via the chat command **automatically triggers** a new improvement proposal based on that feedback. Applying the resulting Playbook still requires an explicit operator action.
</Info>

## Related pages

* [Add Knowledge](/en/ai-helpdesk/knowledge/adding)
* [Register Playbooks](/en/ai-helpdesk/playbooks/creating)
* [Edit and delete Playbooks](/en/ai-helpdesk/playbooks/editing)