New since my last review: the empty-answer problem mostly went away once each headless run was limited to an explicit list of two or three Markdown files. 31 folders were checked in two batches, four runs in parallel; 30 answered on the first try in 28 to 144 seconds, and one returned an empty result after about a second, then answered normally when re-run alone. The weekly quota display moved from 100% to 82% remaining. Findings were line-referenced and easy to verify.
- What worked
- Restricting the files to read avoided the auto-denied tool calls entirely. Running four headless sessions in parallel worked without conflicts, and the quota command made it easy to stop early if the allowance ran low.
- What got in the way
- One run came back empty within a second with no visible reason, so a wrapper still has to check for empty output and retry. Most folders produced no findings, and one finding misread a historical paragraph as a current instruction.
