GitHub Models introduces AI-powered system prompt enhancement [GA]

You can now use the new “Improve Prompt” button next to your system prompt in the Models Playground.

A system prompt is a predefined instruction or guideline that sets the behavior and tone for an AI model, helping it respond in a specific way to user inputs. This AI-powered tool will refine and optimize your prompt to help you get the best possible results from your chosen model.

Prompt improvements

GitHub Models makes it easy for every developer on GitHub to build AI features and products. Easily try, compare, and implement models in your code for free via the playground or API.

Try the “Improve Prompt” feature in the playground, learn more about GitHub Models, or join the conversation in our community discussions.

Find and fix Actions workflows vulnerabilities with CodeQL (Public Preview)

You can now enable code scanning in your GitHub Actions workflow files. By opting-in to this feature, you can enhance the security of repositories using GitHub Actions.
Actions analysis support includes a set of CodeQL queries developed by the GitHub Security Lab to capture common misconfigurations of workflow files that can lead to security vulnerabilities. You can now easily run these queries as part of Code Scanning’s default or advanced setup and use Copilot Autofix to get remediation suggestions on your findings.

You can opt-in to the public preview by selecting the “GitHub Actions” language via code scanning default setup, or by adding the actions language to your existing advanced setup. New repositories onboarding to default setup after today will start analyzing Actions workflows right away. Existing repositories will not be automatically opted-in as part of the public preview.

Learn more about configuring default setup for code scanning, securing your use of Actions, and vulnerabilities identified with CodeQL.

See more

Copilot Autofix can now be generated with the REST API (Public Preview)

New REST API endpoints for code scanning allow you to request the generation of Copilot Autofix for code scanning alerts. These endpoints also provide the Autofix generation status, along with metadata and AI-generated descriptions for the fixes, and enable you to apply Autofix to a branch. This functionality can be particularly useful for addressing security vulnerabilities programmatically and for tracking the status of alerts with Copilot Autofixes in your system.

To generate Copilot Autofix, call the POST /repos/{owner}/{repo}/code-scanning/alerts/{number}/autofix endpoint.
Additionally, you can retrieve the Autofix and commit it by using the GET /repos/{owner}/{repo}/code-scanning/alerts/{number}/autofix endpoint followed by POST /repos/{owner}/{repo}/code-scanning/alerts/{number}/autofix/commits.

For more information, see: About Copilot Autofix for CodeQL code scanning. If you have feedback for Copilot Autofix for code scanning, please join the discussion here.

See more
View all changes