v2.22.0
Cheaper runs with test and verify effort dials
Trau costs less to run by default: how hard agents test and how hard verification hunts are now settings, both shipping at a cheaper level. Settings also gains a machine-wide editor.
Added
- Verification strictness is a dial from rubric-only to a full adversarial hunt, and the rubric still blocks at every level
- The All repos scope in Settings is a real editor, so a value you want everywhere is set once
- Every launch gesture confirms the steps it is about to run before it starts
Changed
- Agents write happy-path tests by default, and a higher test effort setting restores edge-case and error-path coverage
- A running item on the Loop offers Stop, which parks the run at its checkpoint instead of discarding it
Fixed
- The web UI reloads itself after an update instead of getting stuck on a stale version