The architecture of
reliability for CI/CD

Failure intake

OpenSRE picks up failed workflow runs from GitHub Actions and normalises each one into a single investigation state.

Context assembly

OpenSRE enriches the failure with:

• The triggering commit and diff
• Recent workflow and config changes
• Run history and baselines for the workflow

Execute actions

OpenSRE decides whether the failure is CI-caused, code-caused, or unresolved, and identifies plausible causes and investigation objectives.

Agentic investigation loop

• Plans next checks (logs, re-runs, diffs)
• Executes against your repo and CI tooling
• Synthesises evidence into evolving hypotheses

Confidence gate

OpenSRE continuously evaluates:

• Hypothesis confidence
• Whether the fix is verified
• Marginal value of further investigation

Stops and asks you when it isn't sure, or when more checks are unlikely to change the outcome.