Review and approve Copilot agent actions on issues (Public Preview)

[alondahari]

When you set up a Copilot agent automation or Agentic Workflows, the agent can now explain why it made each change and, optionally, ask for your approval before changes take effect.

issues.agentic.review.mp4

Note

This is a preview feature and is experimental. Behavior and availability may change.

What's new

• Rationale: Every action the agent takes (setting type, adding labels, filling fields) includes a short explanation of why. Visible in the Agent suggestions panel on the issue page and via the ⓘ icon on timeline events.
• Suggestions: Instead of applying changes directly, the agent can propose them as pending suggestions. You approve or reject from the issue page, individually or in bulk.
• Confidence: The agent reports how certain it is about each suggestion (LOW, MEDIUM, HIGH). You can set a confidence threshold in your repo's settings to automatically route lower confidence requests to become suggestions, and higher confidence to be automatically applied.

Currently supported for issue type, labels, issue fields, closing issues and assigning users / agents. We plan to expand to additional actions in the future.

Setup (Copilot cloud agent)

1. In your repository, go to Agents → Automations
2. Click Create New and choose the trigger Issue opened
3. Add a prompt:
   • To apply changes directly: "Triage this issue. Set labels and issue type."
   • To require approval first: "Triage this issue and suggest labels and issue type for approval."
4. Enable the List issue fields, Set issue fields, and Set issue type tools.

This is one example of an automation you can create. You can use any of the supported tools and prompt the agent to create powerful automations!

Rationale is always included automatically. To require approval, include the word "suggest" in your prompt. Low-confidence actions are routed to approval regardless.

Feedback

This is a public preview. Please share feedback in this discussion. If you would like to opt out, please post in this discussion or make the change in your repo settings.

[BALANIVEDHITHA]

@alondahari , can this feature be a native github feature when an issue is created .Can it be a native feature to give suggestion on type and priority instead of just getting this when we use agents explicitly ?

[riporipo223]

+1 this would be great as a native github feature basic suggestions like issue type and priority should appear automatically when an issue is created, without requiring a copilot agent, agents can remain for more advanced, customizable workflows

[ankush850]

This is a great preview update for issue management! The review and approval layer makes Copilot agent automations significantly safer and more practical for production repositories.

Here is a detailed breakdown of how to utilize this feature effectively, along with feedback on the workflow:

---

1. Recommended Setup for Automatic Issue Triage

For maintainers looking to set up automatic, native-feeling issue triage on issue creation:

1. Navigate to Settings → Agents → Automations in your repository.
2. Click Create New and choose Issue opened as the trigger.
3. Enable the core issue management tools:
   • List issue fields
   • Set issue fields
   • Set issue type
4. Prompt Configuration:
   • Require Approval (Suggestions Panel): Include the keyword "suggest" in your prompt:

     "Triage this issue upon creation. Suggest appropriate issue type, priority, impact, effort, and labels based on the issue content."

   • Direct Application (Auto-apply High Confidence): Omit "suggest" in the prompt:

     "Triage this issue upon creation. Set labels, issue type, and relevant fields."

---

2. How the Approval & Rationale System Operates

• Agent Actions Banner: When an issue is opened (e.g., #436 Feature Request: Dark Mode Support), an interactive Agent actions bar appears at the top of the issue. Maintainers can preview pending pill suggestions (like Priority → P3, Labels → enhancement, Type → Feature, Effort → High, Impact → Low) and accept/reject them individually or in bulk.
• Inline Rationale (ⓘ icon): In the event timeline, github-actions records each property update with an ⓘ info icon. Clicking it opens the exact reasoning behind the decision.
• Automated Triage Summary: The agent posts a formatted Triage Report comment featuring an Action | Value | Confidence | Rationale table for full auditability.
• Confidence-Based Routing:
  • HIGH Confidence → Applied directly (if configured without "suggest").
  • LOW / MEDIUM Confidence → Automatically routed to pending suggestions regardless of prompt wording, protecting the repo from ambiguous classifications.

---

3. Key Advantage: Agentic vs. Static Native Rules

Static built-in triage rules usually struggle beyond basic bug or feature tags. Copilot Agent Automations excel because they understand custom repository schemas (custom fields like Effort, Impact, custom Priority scales, and project board fields) and evaluate the full technical context of the issue description before assigning values.

---

4. Feedback & Feature Requests for the Team

• 1-Click Automation Templates: Adding a pre-configured template under Automations named "Auto-triage New Issues" with pre-filled tools (Set issue fields, Set issue type) and prompt defaults would make setup even smoother.
• Expanded Action Support: Looking forward to seeing support expanded to Milestone assignment, Project Board column moves, and Duplicate issue detection.

Awesome work by the Copilot team on this preview.

[donjo]

I came across the label suggestions things today for the first time in the wild and I really liked the idea, however, I feel like they would be better presented in a section closer to the place the suggestions would be applied. If the suggestions were for things other than labeling (like an update to the description itself, adding artifacts, etc.) then the current place would make sense to me. Just my 2️⃣ 🪙