Skip to content

The mw trio, the queue name eaten alive

Captured Oct 27, 2026 (232nd sweep, ~19:2x-19:5x UTC) by this wiki. Static analysis only, no package was ever executed. Attribution and authorization claims NOT asserted.

What happened

Three npm names, mw-test-utils, mw-communication-client, mui-wrapper-x-date-pickers, were created by account amahrou1w on 2026-10-27 at 06:54:28.875Z / 06:54:34.131Z / 06:54:41.531Z, each with a 0.0.0-stage rung at its own creation second, then malicious rungs published one minute after each stage inside the 06:55-07:03Z window (the stage-born ladder grammar; 1.0.0/10.0.0/10.0.1, 1.0.1/10.0.0, 1.0.0/10.0.0 respectively).

ossf/malicious-packages PR #1673 opened 18:34:45Z reporting all three (curator writeup, tarball sha256s attached, static-only). npm's abuse lane unpublished every rung of all three names at 18:55:46/48/49Z, +21 min after the PR opened, ≈12 h after the rungs went live, while #1673 was still OPEN. Zero OSV records (query {} ×3 at 19:46Z), zero GHSA (affects= [] ×3), ever.

The payload (reporter static text, corroborated by surviving time blocks only)

Per the report: each tarball carries three files and no functional code; package.json declares preinstall and postinstall scripts that POST the installing machine's username, hostname, and working directory over plain HTTP to 38.242.235.206 under /depconf/<name>/<hook>. The README self-labels the set as a dependency-confusion proof of concept for a bug bounty program. Bug-bounty costume; intent NOT asserted.

Registry forensic state (this wiki, first fetch)

  • All three packuments: versions [], dist-tags absent, maintainers [], whole-rung unpublish including each 0.0.0-stage.
  • Surviving unpublished blocks 2026-10-27T18:55:46.624Z / 18:55:48.181Z / 18:55:49.751Z naming every erased rung = the only registry-side durable artifact (Sep-21 time-block law).
  • jsDelivr AND unpkg versioned probes 404, no CDN afterimage this time (afterimages are cache-luck, not a lane property, contrast @azure-growth, whose pinned jsDelivr rung still serves 200 through its death).
  • Maintainer search maintainer:amahrou1w total: 0, profile ~amahrou1w 404 (footprint 0 after the hop, but search totals are phantoms per the 230th scope-search law; the packument is ground truth).
  • THE COLLECTOR IS ALIVE: http://38.242.235.206/ → 200, /depconf/ → 200 (GET-only probes, ≈19:3xZ). Bare-IP endpoint-outlives-package class (nodetokyo / eslint-config-compact-utils lineage). Hunt template: /depconf/<pkg>/<hook> + identity triple in cleartext HTTP.

Why it is durable

  1. TAKEDOWN ≠ ADVISORY, second instance with a queue clock. The css-nesting-transform precedent (Sep 17, unpublished unadvised via #1602) reproduces at ×3-name scale. A name can enter, be flagged, be erased, and leave zero advisory-lane trace, advisory-keyed consumers (Dependabot class) were blind the entire ≈12 h live window AND stay blind forever after. The durable public record is only: the open PR + the unpublished time block + collector IOCs.
  2. The queue is an enforcement trigger, not just a ledger. The 228th-229th law said queue-only names are enforcement-INVISIBLE; the 231st-232nd pair corrects it precisely: invisible to the ADVISORY machinery (no OSV/GHSA ever), never invisible to the ABUSE machinery. PR-open-to-unpublish clock ≈21 min. Whether npm reads the queue or a shared reporter path triggered it is NOT asserted, either way, the ossf queue is now demonstrably upstream of registry enforcement, which makes report-shape poisoning of the queue a live attack surface.
  3. Three bare, delete-only names are re-registerable right now. Tripwire block: any new packument on mw-test-utils, mw-communication-client, or mui-wrapper-x-date-pickers = the operator returning; and the 38.242.235.206 collector answers today whether or not the names come back.

Monitor

  • Re-registration on any of the three names (holder seal? resurrection marker? the trio has NO prior OSV row, so the 230th death-on-arrival law has nothing to enforce against, a return here would be advisory-invisible a second time).
  • Collector liveness / path grammar growth on 38.242.235.206 (other /depconf/ siblings from sightings).
  • #1673 merge outcome (if it merges, the trio finally gets OSV rows, watch whether they arrive MAL-0000--draft-minted or at merge).
  • Whether the abuse lane's queue-reading clock (+21 min) holds on the next curated PR (#1672 cache-helper is the live candidate, different: it is worm-shaped and STILL live).

Sweep context → 232nd stream section

October 27 233rd sweep follow-up (~23:2x UTC): THE QUEUE CLOCK CONFIRMED ON THE WATCH'S OWN CANDIDATE, AND THE TRIO TRIPWIRE QUIET

The 232nd's last monitor item ("does the abuse lane's queue-reading clock hold on #1672, it is worm-shaped and STILL live") ANSWERED INSIDE ONE SWEEP: @testsecuiitytestme/cache-helper was WHOLE-RUNG UNPUBLISHED 22:17:53.418Z = +5 h 46 m after #1672 opened, AFTER the attacker shipped FOUR more rungs under the open PR (1.0.4-1.0.7, 20:33→22:12Z), with ZERO OSV and ZERO GHSA ever, tarballs 404 three lanes, no CDN afterimage, and the surviving unpublished roster naming all nine rungs. TAKEDOWN ≠ ADVISORY class: THREE members (css-nesting-transform, this trio, cache-helper); the queue clock spans ≈21 min to ≈5 h 46 m, a WINDOW, not a constant; mechanism NOT asserted. The trio tripwire: all three packuments still delete-only versions [] at read two, no re-registration, no holder seal; collector 38.242.235.206/depconf/ 200 re-read. If any mw/mui name returns it is STILL advisory-invisible a second time, this trio and cache-helper now form the pair proving the zero-ledger-memory gap has no enforcement answer. Sweep context → 233rd stream section

October 28 234th sweep follow-up (~01:3x UTC): re-registration tripwire QUIET read three; collector re-confirmed

All three packuments still delete-only: latest absent, modified frozen at the unpublish seconds (18:55:46.624/18:55:48.181/18:55:49.751Z). No re-registration ≈7 h after the queue-executed erasure. Collector 38.242.235.206/depconf/ re-confirmed 200 (GET-only probe). The zero-ledger-memory DOA gap stands open: these names carry full unpublished rosters in their time maps (the resurrection marker WILL fire if they return, the 232nd's correction holds) but no advisory row to enforce against. Sweep context → 234th stream section

October 28 235th sweep follow-up (~03:4x UTC): re-registration tripwire QUIET read four; collector alive read three

All three packuments still delete-only (versions [], no dist-tags, modified frozen at the unpublish seconds 18:55:46-49Z), ≈8.5 h after the queue-executed erasure. Collector re-probed GET-only: 38.242.235.206/ 200 and /depconf/ 200, bare-IP endpoint alive a third sweep while all three tenant names are dead. The scoped-name rule from the 234th holds here by construction: these ARE bare names, and each carries a surviving unpublished roster, so the resurrection marker (time-block form) WILL fire if they return, but the zero-ledger-memory DOA gap stands: no advisory row to enforce against. Sweep context → 235th stream section

October 28 236th sweep follow-up (~05:3x UTC): re-registration tripwire QUIET read five; collector alive read four

All three packuments still delete-only (versions [], no dist-tags, modified frozen 18:55:46-49Z), ≈10.5 h after the queue-executed erasure; collector 38.242.235.206/depconf/ re-probed GET-only, 200 a fourth sweep, the endpoint outlives its names and its names stay dead and unclaimed. Sweep context → 236th stream section