T7 edge cases: every row of the edge-case matrix is a test
main·
Commit

147 offline tests pass and the live rows pass with Docker. Planner stops with PlanError instead of guessing when the model fails, and caps runbooks at 10 steps. Docker down, auth failure, corrupt precedents and network drops each end with a clear report. Secrets are redacted from step output. Verification blocks every model call. Patch reasons end on a whole sentence. Test runs never export traces to neatlogs.
Entire-Checkpoint: 01M4JDQF7XYGJW7R9ZZJKC2TNT
Checkpoints
Checkpoint 1