Test evidence on every proposal
Instead of trusting a diff on faith, every proposal carries the results of your actual test suite, run against the actual change, in the sandbox it was written in.
How it works
TestAgent runs the project's configured test command — the same one your CI would run — inside the proposal's sandbox.
The agent's changes are applied first, so the results reflect the change under review, not a stale baseline.
Failing test names and a regressions-vs-main comparison are attached, so a human sees exactly what broke, not just a pass/fail count.
TestAgent is a system executor, not an LLM, and it never blocks a proposal — the human weighs the evidence and decides.
The agents behind it
Human control
Proposal gate — test evidence is attached alongside review and security findings; only a human decision moves the proposal.