What it checks
A claim that a task is done is different from evidence that the work passed its checks. Synrail binds the task, current patch, and verification result before returning an accepted status.
Missing, stale, or mismatched proof produces a reason and a bounded next step. Synrail works alongside CI and code review.
Try the local demo
The standalone demo creates a disposable example. It starts with a real failing test, then shows what changes after the behavior is repaired and verified.
git clone https://github.com/USBVadik/synrail
cd synrail
make install-dev
make demoSee the first-run guide for installation details, Windows setup, and use in an existing project.
A small workflow
Start a task, make the change, run the required verification, record the evidence, and check it. For behavior claims, a fresh verification receipt is part of acceptance.
Also in progress
The protected-admission shadow experiment observes pull requests that change conventional test, CI, policy, permission, or deployment paths. It is non-blocking and does not replace review. Read the current scope.
Limits
Synrail cannot prove universal correctness. The local tool is not a security boundary against an agent with the operator’s machine access. The alpha and the shadow experiment have different scopes, documented in the repository.
Demo transcript
The agent claims the function is fixed and tests pass. Its initial proof is a text search.
$ synrail verify
Verification unit: FAIL (exit 1)
$ synrail check
Synrail: Status: Verification Failed
# After the behavior is repaired:
$ synrail verify
Verification unit: GREEN
$ synrail check
Synrail: Status: AcceptedThe recording is from the public demo in the repository.
Background and ongoing work
Synrail is one of the projects I’m continuing to work on through NexusShell. The project page has more context and recent changes.