참고
Rationale, confidence, and approvals for automations are in 공개 미리 보기 and subject to change.
Automations run an agent to triage issues for you, applying labels, setting types and fields, assigning issues, and closing them, without a person initiating each change. Rationale, confidence, and approvals give you visibility into that automated work and control over when it applies.
This capability is not specific to any one platform. It works the same whether the agent runs through Copilot 클라우드 에이전트 automations, GitHub 에이전트형 워크플로, or the REST and GraphQL APIs.
This capability applies only to the changes an 자동화 makes to issues. It does not apply to changes people make by hand, and it does not affect other actions an 자동화 can take, such as opening pull requests or pushing code.
Rationale, confidence, and approvals cover changes an 자동화 makes to the following issue attributes: labels, fields, issue type, closing issues, and assignees.
Rationale
Every supported action records the reason behind it, whether the 자동화 applied the change automatically or proposed it as a suggestion. This gives you an audit trail of what changed and why. When a change is waiting for review, you see the reason on the suggestion before you decide.
You can also view the rationale on the issue for changes that took effect, whether they were applied automatically or suggested and then approved. An action with a recorded rationale will have next to it. Select the icon to reveal the 자동화's reasoning for that change.
Confidence
For each supported action, the 자동화 rates its confidence as high, medium, or low. Your repository has an automation level that sets the confidence threshold: changes rated below it are held as suggestions, and changes at or above it apply automatically.
- Full control: Every change is held for review. Nothing is applied automatically.
- Cautious (default): Only high-confidence changes are applied automatically. Everything else is held for your review.
- Balanced: Routine, clear-cut changes are applied automatically. Anything with ambiguity is held for review.
- Full automation: Every change is applied automatically. The agent only holds a change back if it's flagged as uncertain.
Because lower-confidence actions are the ones held back, you spend time only on the changes most likely to need a second look.
Approvals
Whether a change applies directly or waits for your review depends on two things:
- Your repository's automation level. Changes rated below the threshold are always held as suggestions, no matter what you prompted the 자동화 to do.
- An explicit request to suggest. You can also prompt the 자동화 to suggest a change rather than apply it, and it waits for your review even if its confidence is high.
Suggestions appear in a panel on the issue. From the panel you can:
- Accept or decline each suggestion individually.
- Accept all or decline all to act on multiple suggestions at once.
A small team moving quickly might choose Full automation, while a busy public repository can choose Cautious or Full control to hold more changes for review where the stakes are higher.
참고
Approvals are a workflow convenience, not a security control. They don't enforce a server-side boundary, so an agent with permission to change issues can apply changes directly instead of proposing them, including through the REST and GraphQL APIs. Use repository and agent permissions, not approvals, to control what an 자동화 is allowed to do.
Next steps
To learn how to configure your repository's automation level and manage suggestions, see Managing rationale, confidence, and approvals for issues.