Meet BotHard.

The repair bot. While your agent writes the code, BotHard is down on the shop floor, catching what fails and sending it back for another try, for up to ten attempts. BotHard is our name for the factory’s repair loop.

He takes the hit.

When a gate fails during a build, BotHard collects the full failure output (both output streams) and hands it to your agent with the exact check that failed.

[gate] sast: blocking finding in src/db.ts
[bothard] failure sent to your agent with both output streams
[agent] patch submitted
[gate] re-running all gates...
[gate] chain passes

Illustration.

He stays in the loop.

BotHard takes the patch and re-runs every gate, for up to ten attempts. A fix only counts when the whole chain passes.

He works in a sandbox.

Generated code is built and tested in your operating system’s sandbox. On Linux there is no network except for dependency installs; on Mac the sandbox blocks network and limits writes but does not yet isolate reads. If no sandbox is available, the build says so and runs unsandboxed.

Coming soon

For hosted workflows, BotHard will propose production fixes for your approval; fixes will not be deployed automatically.