Inspiration

Release decisions often happen across CI logs, test results, and rushed team messages. Teams can see that something failed, but not clearly answer: “Can we ship, why, and who approved the risk?”

Release Guardian is being built to make that decision visible and accountable.

What it does

It turns bounded CI evidence into one release assessment:

  • What passed or failed
  • Which release policy is affected
  • The exact evidence behind the result
  • A proposed remediation
  • An explicit human approval, rejection, or time-limited exception

It never merges code or deploys automatically.

How we’re building it

The core is TypeScript with deterministic policy evaluation and tested CI-evidence collection. GitLab CI provides the release signals. GitLab Duo Agent Platform supports the project workflow through repository instructions and review rules.

The dashboard will show the evidence timeline, policy gate, remediation proposal, and append-only decision record.

Challenges

The hard part is useful AI without unsafe authority. We are solving that by separating evidence, policy evaluation, proposals, and human decisions. A recommendation is never treated as permission to ship.

What we learned

Release automation is more trustworthy when every conclusion is traceable. The product should explain uncertainty and block unsupported claims instead of pretending to know.

Built With

Share this project:

Updates

Submission history