Most teams (and agents) can produce diffs far faster than they can validate them. The bottleneck is not writing code — it’s rebasing, reviewing, testing, documenting, and making a branch safe to share. This tool moves that bottleneck into a deliberate local gate: push to a special remote and an isolated pipeline readies the change before anything becomes public.
What Sets It Apart
- Disposable worktree and non-blocking pipeline — runs the full validation flow in an isolated worktree so your working directory is never rewritten. So what: you can keep coding while the gate prepares a clean branch for upstream.
- Agent-agnostic but agent-native — plugs into many LLM/agent providers and also exposes a
/no-mistakesskill so agents can drive the same approval flow. So what: you can gate agent-authored changes headlessly or use the TUI to review findings interactively. - Safe auto-fixes plus human approval points — mechanical fixes are applied automatically; anything that touches intent pauses for you to approve, fix, or skip. So what: reduces trivial back-and-forth while ensuring human control over substantive intent changes.
- End-to-end push → PR → CI handling — when checks pass the gate pushes to the configured target and opens a PR, then watches CI and attempts auto-fixes for CI/merge issues. So what: it turns a fragile manual push workflow into a reproducible, reviewable pipeline.
Who It's For & Trade-offs
Great fit if you want to enforce a reproducible, pre-push validation flow (manual or agent-driven) without blocking your local worktree, or if you run agents that produce many rapid changes and need an automated gate. Look elsewhere if you need a lightweight pre-commit hook only (this is a fuller pipeline that runs in a separate worktree), if you cannot accept sending code/metadata to external agent providers, or if integrating agent credentials into developer machines is unacceptable for your security policy. It also adds operational complexity compared with plain git pushes: initial setup and per-repo configuration are required.
How it typically fits in a workflow
Developers or agents push to the special remote, the gate performs ordered steps (intent → rebase → review → test → docs → lint → push → PR → CI) in isolation, applies safe fixes, surfaces findings for human decisions, and only forwards the branch when everything is green. The net result is cleaner PRs and fewer manual remediation loops after a public push.
