GitHub Issues public preview adds agent automation controls

New capabilities include Approvals, Confidence, and Rationale for agent automations in GitHub Issues. Approvals allow suggestions to be reviewed before application, Confidence rates actions for automatic application or review, and Rationale provides an audit trail for each change.

Rationale, confidence, and approvals are integrated with GitHub Agentic Workflows and Copilot cloud agent automations. These features are also accessible through the REST and GraphQL APIs, covering changes to labels, fields, type, close, and assignees.

Repository administrators can set confidence thresholds to control which automated changes are applied directly and which are held for review. This allows for customization of automation levels based on repository needs, such as allowing full automation for fast-moving teams or review steps for busy public repositories.

Existing GitHub Agentic Workflows can be upgraded to add issue intent support, remaining backward compatible. Users can require intents for specific safe outputs by setting 'issue-intents: true' in their workflow frontmatter or opt out with 'issue-intents: false'.

Potential applications include automatically triaging issues with reasoning provided and allowing optional review, as well as enriching issues with missing metadata like labels or types. Spam detection is also facilitated by flagging suspected spam with reasons and holding uncertain cases for review.