Automatically resolve stale or already fixed GitHub issues and PRs powered by AI.
Continuous AI Resolver is a GitHub Action that runs weekly (or on demand) to help maintainers keep their repos tidy. It uses an LLM to check open issues and pull requests and determine if they’ve already been addressed in the codebase. If so, it comments with context and can optionally close them.
Maintainers spend too much time triaging duplicates, stale reports, or PRs that no longer apply. This agent runs in the background and acts as a second set of eyes, saving time and surfacing what’s already handled.
- Detects stale or resolved issues and PRs using LLM context matching
- Posts a comment with explanation and links (e.g., which file/function fixed it)
- Checks for missing contribution info (like steps to reproduce)
- Optionally labels incomplete issues with needs-info
- Re-runs weekly to catch newly fixed items
The resolver:
- Loads open issues and pull requests for your repository
- Scrapes their text and matches against code context
- Uses OpenAI (GPT-4) to determine if the thread is already resolved
- Posts a smart comment, if applicable
- Skips threads already if reviewed
- Add this workflow to .github/workflows/continuous-ai-resolver.yml (already included in this repo)
- Create an OPENAI_API_KEY with access to GPT-4. Optionally, you can self-host a model and point to it via OPENAI_BASE_URL.
- Customize the behavior in main.js, openai-client.js, or resolver.js.
- (Optional) Adjust your repository’s CONTRIBUTING.md — it’s auto-referenced when checking for issue completeness.
Or run it manually:
- You can trigger the workflow manually via the Actions tab (choose Run workflow) or wait for it to run on Sunday at midnight UTC.
If an issue is resolved or missing required details:
- Post a structured comment (including a link to the resolving code)
- Optionally apply a
needs-infolabel - Skip duplicates already commented on
This is built to be forked and improved.
MIT with ❤️ for open source maintainers.