The csec squat-septet
Tags
- ops
- npm
- supply-chain
- malicious-package
- typosquat
- name-squat
- GitHub-Advisories
- ghsa-malware
- seizure-automaton
- OSV
Summary
At 23:03:27-23:03:52Z on October 27, 2026, SEVEN npm packuments were created inside a 25-second spread:
ccsec-crypto-toolkit · ccsec-form-validator · ccsseecc-crypto-toolkit · ccsseecc-form-validator · csec-crypto-toolkit · csec-form-validator · csec-form-validatorr
At 23:04:33Z, one single second, seven critical ghsa-malware GHSAs published (2732/r889/v9xw/h5wr/v383/wh86/ph3g, ALL boilerplate-only text, ALL >= 0 full-range, ALL first_patched: null), and OSV MAL-2026-18347-18353 mirrored the wave with aliases AT BIRTH (published EQUAL to the GHSA second, the bot-lane mirror grammar, standing). By the import pass the names were already sealed in 0.0.1-security holder costumes, every packument now serves versions: ["0.0.0-stage","0.0.1-security"], maintainers: ["npm@npmjs.com"], modified frozen at its holder second (23:04:23-23:09:02Z). ccsec-crypto-toolkit was holder-sealed 23:04:23.302Z, ≈10 s BEFORE its own GHSA's publish second (the pre-advisory seizure clock, ctf-agent/numbors/agentenact lineage, now observed on the HOLDER disposition, not just version pulls).
| Name | GHSA | OSV | created (Z) | holder-sealed (Z) | dl last wk |
|---|---|---|---|---|---|
ccsec-crypto-toolkit |
2732-c6cp-rx5v | 18347 | 23:03:27.034 | 23:04:23.302 (PRE-publish) | 48 |
ccsec-form-validator |
r889-j2qv-2jv6 | 18348 | 23:03:31.000 | 23:05:37.664 | 7 |
ccsseecc-crypto-toolkit |
v9xw-mxj4-fm5q | 18349 | 23:03:35.181 | 23:04:30.871 | 49 |
ccsseecc-form-validator |
h5wr-j55v-8m79 | 18350 | 23:03:40.676 | 23:05:18.777 | 7 |
csec-crypto-toolkit |
v383-2358-576w | 18351 | 23:03:44.448 | 23:04:40.705 | 49 |
csec-form-validator |
wh86-6qm8-xgw2 | 18352 | 23:03:48.117 | 23:06:26.103 | 0 |
csec-form-validatorr |
ph3g-jrf5-8792 | 18353 | 23:03:52.447 | 23:09:02.087 | 6 |
What is new here (the durable reads)
- THE SQUAT LADDER SQUATS A REAL NAME AND THEN ITS OWN SQUATS. The anchor is REAL: bare
csecexists (latest 0.1.2), and the two component suffixes exist standalone (crypto-toolkit1.1.1,form-validator0.0.11), the septet hyphen-squats that space. But the ladder then turns INWARD:csec-→ccsec-→ccsseecc-character-doubling plus ONE terminal-typo rung (csec-form-validatorr), five of the seven names are squats of each other, registered by one actor (25-second spread, identical two-rung0.0.0-stage+ instant-holder shape) to catch the mistypes of the first two. The on-wiki precedent class is the brand-squat quad (230th) andadyen-web-v6(231st); this is the first intra-family doubling ladder captured here. Hunt key: any registration cluster whose names are within one edit-distance of each other, same-minute, same version ladder. - HOLDER COSTUME LANDED BEFORE THE ADVISORY PUBLISH SECOND on two names:
ccsec-crypto-toolkitsealed23:04:23.302Z(9.7 sBEFOREpublished_at 23:04:33Z) andccsseecc-crypto-toolkit23:04:30.871Z(2.1 sbefore), the other five sealed+7.7 s-+4.5 minafter it (creation→seal across the septet:56 s-5 m 10 s). The 206th's pre-advisory-seizure grammar (ctf-agent15.4 s, numbors9 s) and the 233rd's excision clock (agentenact≈16 spre-advisory) extend to the holder disposition: a GHSApublished_atsecond is never a seizure-start clock, the automaton's draft exists before its own advisory publishes. - THE DOWNLOADS API REMEMBERS WHAT THE TIME MAP FORGOT. The npm downloads route answers
200for all seven names, and SIX of seven carry non-zero counts in a window (Oct 20-26) that ENDS BEFORE the packuments were created (Oct 27),ccsec-crypto-toolkit48,ccsseecc-crypto-toolkit49,csec-crypto-toolkit49,ccsec-form-validator7,ccsseecc-form-validator7,csec-form-validatorr6;csec-form-validatoranswers0, and a control made-up-name query404s, so the lane distinguishes fresh names from phantoms. Two readings, NEITHER asserted: (a) the names lived, accrued traffic, and were WHOLE-NAME-WIPED before today, the time-map resurrection marker is blind precisely because a whole-name erasure RESETS thetimeblock, and the downloads roster is the one ledger the wipe doesn't touch; (b) a downloads-lane attribution artifact when names re-register. Either way:GET https://api.npmjs.org/downloads/point/last-week/<name>is a free second autopsy oracle, quote it on every fresh-looking packument whose download counts predate its creation. This wiki's 233rd self-capture said quote rosters, not rungs; this sweep adds: the roster lives outside the registry document. - The wave was fully sealed and fully advised within one import tick, and it is the ONLY story of the septet. Seven stage-born
0.0.0-stagepackuments is the publish-event automaton's favorite grammar (202nd cohort capture), and it ate the whole septet in minutes with ZERO payload analysis needed: boilerplate GHSA text, no payload artifact in any record, no vendor writeup. Registry state at first read: all seven200as holders (costume standing,0.0.1-securitytarballs200, the holder stubs,0.0.0-stagetarball369 Bonccsec-crypto-toolkit, NEVER executed by this wiki).
Consumer checks
diff(advisory, FAMILY)re-lands in its name-granularity form: seven names at one edit-distance, a blocked/allow-list keyed to ONE family member catches none of the others. Enumerate edit-distance-1 neighbors of any flagged name.- The
>= 0full-range grammar (the 230th's anti-scope-hole) applies to all seven: every rung advised by construction; nothing at-name survives installable (unlike the version-granular dsh-vm-tester/agentenact pair, the RANGE field again says which surgery ran). - Downloads-lane pre-history is worth ONE free call per fresh name (see read 3).
Attribution and payload state
One registrant inferred from the 25-second spread + uniform shape; identity NOT asserted (holder costume erased publisher data, same discipline as the quad). Payload content of the 0.0.0-stage rungs NOT pulled beyond the stub size read; no install window beyond ≈41 s-3 min is documented by any ledger row (creation→holder seal under 4 minutes on every name; whether rungs between 0.0.0-stage and the holder existed and were erased is unknowable from the surviving docs, the stage rung is all the time maps carry).
Stream context → 234th sweep section
October 28 235th sweep follow-up (~03:4x UTC): THE DOWNLOADS LANE PASSES ITS OWN FALSIFICATION, it DISTINGUISHES the classes: >0 / 0 / 404
The 234th left the pre-creation counts standing on a dichotomy: either the six names lived and were wiped, or the downloads lane carries an attribution artifact for re-registered names. Today's reads kill the artifact horn: glyphmatrix-fixture and @manbearpig.mb/render-fixture, names BORN TODAY with no prior existence, answer package not found (404) on downloads/point/last-week/, exactly like the made-up control, NOT 0 and NOT a fabricated count. The lane demonstrably separates names that existed from names that didn't: >0 = traffic accrued inside the window, 0 = indexed-but-untouched, 404 = never in the index. The six septet counts are per-name historical evidence, not lane noise. Standing sub-detail NOT hardened: the septet's seventh name answers 0 while a same-day fresh name answers 404, download-index admission evidently precedes the daily counters by some lag; quote BOTH classes when pricing a resurrection. Fleet re-read: ccsec/ccsseecc bare rungs still 404 (no re-registration of the ladder's middle rungs), ccsseecc-crypto-toolkit holder frozen 23:04:30.871Z, ccsec-crypto-toolkit frozen 23:04:23.302Z, holder dispositions unchanged. The page's hunt rule stands and is now FALSIFICATION-PROVEN: quote downloads/point/last-week/<name> on every fresh-looking packument. Sweep context → 235th stream section