Watches production, reads issues, and opens the fix PR.
Libra connects GitHub, OTEL, logs, and deploy history so engineering teams can see what broke, who is affected, which PR likely caused it, and the fastest path to repair.
GitHub-native reviews · runtime-aware deploy checks · issue-to-PR automation
Scheduled agent actions over 15 minutes cannot enqueue
Users may miss delayed automation because SQS rejects DelaySeconds above 900.
- Read issue and generated implementation spec
- Maintainer approved with libra:approved
- Sandbox is opening a focused PR
Libra Issues
Label an issue. Get a PR.
Libra listens to GitHub issues, turns vague bug reports into an implementation spec, waits for maintainer approval, runs a Fly microVM, pushes a branch, and comments back with the pull request.
Apply libra:ready
A GitHub issue webhook creates the job and a spec comment.
Review the spec
Libra asks clarifying questions if the issue is not actionable yet.
Approve
Add libra:approved when a maintainer wants the agent to implement.
PR opens
The sandbox runs GSD/Claude, pushes libra/issues/*, and links the PR back.
runtime-aware deploys
Every deploy gets a verdict tied to real user impact.
Libra compares changed files, merged PRs, traces, CloudWatch logs, ClickHouse aggregates, and known runtime fingerprints so teams can distinguish a real regression from shared environment noise.
Needs attention now
Production failures with plausible current customer or operational impact.
Before promotion
Staging signals that would likely affect users if shipped to prod.
Concerning, not blocking
Security, performance, and ambiguous findings that need review but not a rollback.
Built for production repos.
beta
Start with deploys and issues.
Connect GitHub, install Libra Issues, and let Libra prove value on your real PRs before rolling it out broadly.