Split from #5214.
Problem
Whole-home scans hit the entry cap and there is no automatic split. Repro: run a whole-home inventory with --confirmed-large-scan. It exits 2 with:
snapshot exceeds 250000 entries; rerun with --max-depth or split the audit into bounded subtrees
The cap is MAX_SNAPSHOT_ENTRIES = 250_000 at skills/clean/scripts/hygiene.py:34, raised at hygiene.py:2009. Selecting children with --root-children hits the same cap when the selected children are large. There is no multi-target form and no automatic split, so the worker divided home into four runs by hand and merged the results. --sizes-only returns exact totals but writes no per-entry inventory (entries: 0, no findings), so every audit needs a second pass over each subtree. Each scan took minutes.
Expected
One command that audits a large tree, either by accepting several targets or by splitting on the depth-1 frontier itself and returning a documented set of snapshots.
Suggested direction
Add an --auto-split option (or a documented orchestration helper) that uses the --sizes-only frontier to plan bounded subtree scans, runs them, and reports coverage. At a minimum, make the cap error name the largest children so the caller can choose the split.
Split from #5214.
Problem
Whole-home scans hit the entry cap and there is no automatic split. Repro: run a whole-home inventory with
--confirmed-large-scan. It exits 2 with:The cap is
MAX_SNAPSHOT_ENTRIES = 250_000atskills/clean/scripts/hygiene.py:34, raised athygiene.py:2009. Selecting children with--root-childrenhits the same cap when the selected children are large. There is no multi-target form and no automatic split, so the worker divided home into four runs by hand and merged the results.--sizes-onlyreturns exact totals but writes no per-entry inventory (entries: 0, no findings), so every audit needs a second pass over each subtree. Each scan took minutes.Expected
One command that audits a large tree, either by accepting several targets or by splitting on the depth-1 frontier itself and returning a documented set of snapshots.
Suggested direction
Add an
--auto-splitoption (or a documented orchestration helper) that uses the--sizes-onlyfrontier to plan bounded subtree scans, runs them, and reports coverage. At a minimum, make the cap error name the largest children so the caller can choose the split.