@workglow/sec@0.2.0 was published 2026-09-08T22:27:24Z from 7942d736. The two fix commits
landed the next day:
$ git log --format='%h %ci %s' 2ab82ba..6d9ee18 | tac | tail -3
7942d736 2026-09-08 22:22:41 +0000 chore: release @workglow/sec@0.2.0
532f61d2 2026-09-09 19:30:59 +0100 fix(adv): bound the ingest folder and keep the newest filing per CRD (#363)
6d9ee187 2026-09-09 19:31:19 +0100 fix(kb): honour --dry-run when indexing, and page the chunk scan (#364)
$ npm view @workglow/sec version
0.2.0
So the build an installer gets today still has, by each commit's own description:
IngestAdvSnapshotTask joining SEC_RAW_DATA_FOLDER with a completely unvalidated folder —
"any local *.csv could be read into adv_row" — and snapshot unvalidated when folder is
supplied, so "a chosen snapshot label wiped a real one through the deleteSearch that runs
first". Reachable from sec-base task run, the web console form and the MCP tool surface.
sec index --dry-run and sec ask --dry-run committing kb_document and kb_chunk rows —
after runCommand has printed "Dry run — no data will be written".
SELECT * FROM kb_chunk with no bound per question, hydrating the whole index to score one
query.
Also missing from 0.2.0: the newest-filing-per-CRD reduction, so adv_adviser on the cumulative
archive keeps whichever amendment the CSV happened to list last and sec show advisers --min-aum
filters on an arbitrary historical AUM.
None of these is a regression introduced by 0.2.0 — they are all older — but the release sits
between the review that found them and the commits that fixed them, which is the one window where
cutting a patch costs nothing and buys everything.
24025a60 made this a single command, and bunset --auto will derive patch from two fix:
commits with no break markers:
Suggested as a standing habit rather than a one-off: when a release lands before the fixes for
the defects its own changelog window surfaced, the follow-up patch is part of the fix, not a
separate decision.
Found during the 2026-09-14 review. Snapshot: workglow-dev/prd → analysis/grades/2026-09-14/sec-detailed.md §5.2.
@workglow/sec@0.2.0was published 2026-09-08T22:27:24Z from7942d736. The two fix commitslanded the next day:
So the build an installer gets today still has, by each commit's own description:
IngestAdvSnapshotTaskjoiningSEC_RAW_DATA_FOLDERwith a completely unvalidatedfolder—"any local
*.csvcould be read intoadv_row" — andsnapshotunvalidated whenfolderissupplied, so "a chosen snapshot label wiped a real one through the
deleteSearchthat runsfirst". Reachable from
sec-base task run, the web console form and the MCP tool surface.sec index --dry-runandsec ask --dry-runcommittingkb_documentandkb_chunkrows —after
runCommandhas printed "Dry run — no data will be written".SELECT * FROM kb_chunkwith no bound per question, hydrating the whole index to score onequery.
Also missing from
0.2.0: the newest-filing-per-CRD reduction, soadv_adviseron the cumulativearchive keeps whichever amendment the CSV happened to list last and
sec show advisers --min-aumfilters on an arbitrary historical AUM.
None of these is a regression introduced by
0.2.0— they are all older — but the release sitsbetween the review that found them and the commits that fixed them, which is the one window where
cutting a patch costs nothing and buys everything.
24025a60made this a single command, andbunset --autowill derivepatchfrom twofix:commits with no break markers:
Suggested as a standing habit rather than a one-off: when a release lands before the fixes for
the defects its own changelog window surfaced, the follow-up patch is part of the fix, not a
separate decision.
Found during the 2026-09-14 review. Snapshot:
workglow-dev/prd→analysis/grades/2026-09-14/sec-detailed.md§5.2.