ai-action

Upgrade from 5.0.19 to the latest stable release, 7.0.128, and rebuild the bundled action. Migrate structured generation to with , map the action's input to the SDK's option, and update the Gateway fixtures and snapshots to the v4 protocol. There are no breaking changes to the GitHub Action's public API. Workflows continue using the same , , , , and inputs and the same and outputs. The action continues running on Node 24. SDK method names, result shapes, and Gateway protocol details are implementation changes and require no workflow migration. Validation: Clean install from the public npm lockfile. : all five scenarios pass with 100% coverage. : the same five scenarios pass against ; this check also runs in CI. Both suites assert that the Gateway Authorization header matches the action's API key input. TypeScript check of and against the installed SDK types. Integration and security CI pass. Live Gateway CI is blocked by the repository's invalid secret: the published v2 action also returns HTTP 401 / “Invalid API key provided” in the comparison run. Replace the secret and rerun End-to-End.

Hi Vercel team 👋 great project! It reminds me a lot of https://github.com/anthropics/claude-code-action albeit agnostic to LLM providers, which is great as it gives more flexibility. Hence, I was wondering if you would eventually like to build out by taking inspiration from , yet having the same functionality as the AI SDK? I could imagine various different workflows to e.g., run multiple parallel agents whenever a new GitHub issue is created to try and solve it, then an evaluator agent that checks all solutions or various other automations that this could enable. I suppose, the sky is the limit, and there is a lot of benefit in having something as agnostic and capable as the AI SDK in GitHub Actions. Wanted to check before I create any Pull Requests to see if there's appetite for such features like the new Agent class, tool calling and so on? Thanks