In development · early access
Roadmap
Three phases, in order, without dates. What moves between them depends on what we learn from design partners, and we'll update this page when it does.
Now
Design, early access and design-partner pilots.
- Starting conversations with engineering teams about the agent changes they've had to roll back, and what those cost them.
- Designing the check: the context diff, the suite format, the verdict rules and the pull request report described on this site.
- Early access: finding a few design partners to pilot Rerunner on one real repository each.
Next
Claude Code regression runs with design partners.
- Regression runs in partners' CI: context diff, deterministic checks and base-against-PR runs with Claude Code, reported on the pull request.
- Mined suites: tasks proposed from a team's merged pull requests, kept only if their checks fail before the merge and pass after it.
- A published benchmark of seeded context defects, with detection and false-alarm rates, so you can judge the checks for yourself.
Later
More agents, and deeper answers.
- More agents: Codex, Cursor, Gemini CLI and OpenCode.
- Root cause: restore each changed piece of context and re-run, to name the edit behind a regression.
- Runtime context gateway: deliver the tested version of your agents' context at session start, so what runs is what was tested.
- Agent replay: replay a recorded agent session against a changed setup to see what would have been different.
Not in the first version
On purpose.
- A composite “reliability score”. A single number hides what changed. We report counts and verdicts per task.
- A confidence percentage we haven't calibrated. Until we can show it means something, we won't print it.
- A dashboard or a hosted service. The first version reports where you already look: the pull request.
How we decide
We don't yet know how often changes to an agent's setup cause regressions in practice. That's the most important thing to learn, and design partners are how we'll learn it. If the evidence says the problem is smaller, or different, than we think, the roadmap changes.