v0.34.0 — stop a batch, and tests that bite
This is the release note as published, not a rewrite of it. Release notes are published in the language they were written in.
Stop a running batch
Every still-running task card in the Agent Manager now has its own Stop, plus a Stop all. Until now only the single coding run could be stopped — a parallel batch had to be waited out. Same mechanism as the single run: one stop flag per task, so stopping one card leaves the others running.
Cancellation is cooperative, not an instant kill: a task finishes its current attempt and then halts, because an in-flight model call can't be interrupted. The UI says exactly that.
The Code screen's remaining components are now tested
The suite went 57 → 91 tests. The file viewer, command runner, browser-screenshot panel and live per-edit diff stream previously had none. The new tests target the places where a regression would be dishonest rather than merely ugly:
- a truncated or binary file offers no Edit (saving the shown text would clobber what was never loaded), and a failed save surfaces the error instead of claiming "Saved";
- a screenshot failure shows the real error and no image — never a fake or broken one;
- the live edit stream keeps true edit order, with a re-edited path as its own later step;
- the command runner says each Run is a fresh subprocess and isn't an interactive terminal.
Every honesty guard was verified by mutation testing — breaking the guard in the component makes the matching test fail. A test that can't fail isn't a test.
If you're on v0.33.0, your app will offer this as an in-place update. Installers + pip install -U 'chimera-agent[desktop]' as usual.
Full changelog: https://github.com/brcampidelli/chimera-agent/blob/main/CHANGELOG.md