OX Security
Summary
OX Security (August 25, 2026) identified a fake Cloudflare CAPTCHA phishing campaign distributed through 24 malicious npm packages sharing the exact same HTML page. The packages do not execute code on install, the actor's use of npm is to exploit registry mirrors (unpkg, npmmirror, yarn, tencent, and others) as free, trusted-domain payload storage: mirrors serve package files (including bare index.html) directly on their own domains, so anyone opening a mirror URL sees a fully rendered phishing page on a reputable host. The page presents a fake Cloudflare verification, then redirects through obfuscated JavaScript to attacker-controlled infrastructure. Packages typically reach 50-300 weekly downloads before removal, but mirror copies persist after registry takedowns.
Tags
- ops
- operations
- OX Security
- npm
- npm mirrors
- unpkg
- npmmirror
- ClickFix
- fake CAPTCHA
- fake Cloudflare
- phishing
- payload storage
- Microsoft typosquat
- keyval.org
- trusted-domain abuse
- supply chain
Campaign mechanics
- Each package's content is a single fake Cloudflare verification HTML page with obfuscated JavaScript.
- First wave: the JS redirected to the typosquatted Microsoft domain
login[.]microsofte[.]live(now blacklisted). - Current wave ("keyval new logic"): the JS fetches an encrypted value from the legitimate key-value store
api.keyval.org, decrypts it, and browses to the result. OX observed the value currently resolving to the legitimate ChatGPT website, but the actor can point it at ClickFix pages or any phishing domain on demand, the npm package is the persistent, trusted-domain front end. - Distribution: links to the mirror-hosted pages are shared via social channels; the packages exist to keep the phishing HTML alive on trusted domains even after removal from npm.
Indicators
- IOCs (OX):
login[.]microsofte[.]live,api.keyval.org(legitimate service abused for storage; not itself malicious). - Example mirror URL:
https://unpkg[.]com/ndmxchdjxn2@1.0.0/index.html(rendered fake-CAPTCHA page). - Affected packages (24, published 2026-08-04 → 2026-08-24 UTC):
bgzxcuite2,prezdentkxheiw,egair0810,mnteckets,airdzticket,egypt0811,passport811,vxhjkseuiaqkb,ndmushdkeqe,ndmxchdjxn2,ndmfguyhoxc3,mjsdqwocvn,m2fcsfyjkuxb,m3fdfocdoewn,@worrisome/reutil,testdgdbcsd,tesgfvbncsdbcv,mndsxcusiwlk1,mn2adskhweox,mn3sadkoiewu,mn4xcouzvhus,mbxcnsuwgs1,skxcmwuncbg2,mobiwaefhxc3, "Microsoft Typosquat" family (random-string names, 7 taken down by August 11) plus "keyval new logic" family (8 live at publication).
Defender heuristics
- Treat npm mirror domains (unpkg, npmmirror, yarn, tencent, etc.) as potential phishing hosts when URLs are not part of normal package-download workflows; add mirror URL shapes to phishing detection and URL-reputation pipelines.
- Check proxy/DNS logs for direct
.htmlrequests (especiallyindex.html) to mirror domains, legitimate tooling downloads package tarballs, not bare HTML pages. - Hunt the fake-CAPTCHA primitive in email/browser telemetry: a page claiming browser security verification that offers a "run this command" clipboard action is the ClickFix shape (see the macOS ClickFix fingerprinting-gate page and the ACR Stealer page).
- Abuse of trusted key-value/short-link services (
api.keyval.orgencrypted-value redirection) should be treated as an indicator of phishing redirect infrastructure; monitor outbound resolution of such lookups from user browsers. - Assume persistence after takedown: mirror copies of removed packages survive; hunt for mirror-hosted
index.htmlURLs in social links and phishing reports even when the npm package no longer exists.
Same-day updates
October 21, 2026: the scope RE-ARMED under its own advisory, then the lane closed (this wiki forensics, one-hundred-and-sixty-seventh sweep)
The campaign lane this page was written about came back to life today, on a name that had carried a public OSV record for ≈56 days. Captured live by this wiki at ~21:2x-21:5x UTC Oct 21:
@worrisome/reutil, in OX's original 24-name list,MAL-2026-15604since Aug-26, yet NEVER holder-seized: the name stayed installable withlatestserving (downloads162/wk in the Oct 12-18 window measured by this wiki). Registrytimeblock (read directly): today at02:12:50Zthe publisher shipped a NEW version1.0.5(prior versions1.0.0-1.0.3all date Aug-14) = a re-arm commit landed on an already-advised, still-installable name ≈19 h before enforcement noticed.@worrisome/aaaa, NOT in OX's published 24-name list;1.0.0dates Sep-8 = scope expansion AFTER the campaign post. New1.0.1shipped today02:15:43Z, three minutes after the sibling's re-arm. Downloads4/wk.- Enforcement, one second wide: holder-seizure doc re-creation
21:19:43.7/21:19:52.4Z, holder versions0.0.0-stage+0.0.1-securitythrough21:20:48Z, GHSA pair SAME SECOND21:20:33Z(GHSA-fxf3-w84v-3v6gaaaa,GHSA-4f5p-mp5p-863freutil,>= 0blankets), OSVMAL-2026-18125minted for aaaa with alias-at-birth, andMAL-2026-15604forward-joined to its new GHSA at the21:30:24Zimporter tick ≈10 min after the seizure. - Artifact state: BOTH re-arm tarballs (
1.0.5,1.0.1) are404on the registry and jsDelivr at check = the new-build content is unknowable; jsDelivr still serves the OLDreutil@1.0.3= CDN survivor class. The VERSION EVENT is what the registrytimeblock proves; the payload is recorded-not-executed. - Lane closure: the seven
keyval new logicnames not in tonight's pair (mndsxcusiwlk1,mn2adskhweox,mn3sadkoiewu,mn4xcouzvhus,mbxcnsuwgs1,skxcmwuncbg2,mobiwaefhxc3) re-read holder-seized,modifiedfrozen at the02:42-02:43ZSept-23 wave that skipped reutil, tonight's pair finishes what that wave started,≈58 dafter OX's post.
DURABLE (defender rule): an OSV record is a label, not a lock. Advised package names can be, and here were, re-weaponized with fresh versions while the advisory sits public and the registry keeps serving latest. Consumer-facing detection must key on registry VERSION events for names that already carry advisories, not on advisory presence alone; the bbmarker law (advisory ≠ enforcement ≠ alert) now has its first RE-ARM counter-instance: the unadvised-lane survivors are the re-arm candidates. Cross-stream ledger: one-hundred-and-sixty-seventh sweep section.
October 22, 2026: the jsDelivr reutil@1.0.3 survivor reads 404, survivor-END candidate at seizure+≈40 h, caught behind a scoped-name probe trap (this wiki forensics, one-hundred-and-seventy-fifth sweep)
The CDN survivor re-attested 200 on nine straight sweeps (165th-174th) read 404 this sweep, recorded as an END CANDIDATE, one more scoped read pending, because two clean 404s are state-change evidence while flap class is 000/502. The self-capture is the probe itself: the sweep's FIRST read used the BARE name reutil@1.0.3, which on jsDelivr resolves to a DIFFERENT, legitimate 2024-era package (reutil, maintainer neeswebservice, versions 0.0.2-0.0.5, unscoped root 200), the bare probe measured the wrong package entirely and its 404 carried zero information about the survivor. Only the SCOPED route @worrisome/reutil@1.0.3 addresses the survivor; it read 404 on two independent re-verifications. DURABLE: an unscoped probe of a scoped name is a null measurement, survivor sweeps keyed on package name alone systematically miss (and mis-score) scoped scopes. Registry state holds meanwhile: @worrisome/reutil + /aaaa both frozen [0.0.0-stage, 0.0.1-security], modified 21:20Z Oct-21 = no re-arm at +16 h. Sweep context → 175th stream section.
October 22, 2026: the bbmarker long-hauler is GONE, five-route 404 sweep at ≈8.3 d = the fleet's first silent registry enforcement (this wiki forensics, one-hundred-and-seventy-sixth sweep)
bbmarker, the frozen day-nine name still installing 605 dl/wk as of the Oct-18 capture, answered 404 on EVERY route this sweep: full packument, abbreviated packument, tarball, the downloads API (package bbmarker not found), and jsDelivr, all confirmed by double-read. No advisory, no OSV movement, no notice: the enforcement arrived as a plain deletion ≈8.3 d after the Oct-18 05:50Z freeze. DURABLE: for a fleet name frozen in a security-holder costume, a simultaneous all-route 404 (registry + CDN + analytics) IS the takedown signal, there is no changelog to wait for; and the purge being TOTAL (jsDelivr residue absent) means the CDN-survivor class did NOT catch this name. Re-registration watch open. Sweep context → 176th stream section.
October 22, 2026: RETRACTION, the reutil@1.0.3 survivor is ALIVE: three consecutive scoped reads 200 at seizure+≈44 h; the pair of clean 404s was flap (this wiki forensics, one-hundred-and-seventy-sixth sweep)
The 175th's END-candidate is withdrawn. On ≈15:3xZ re-read the SCOPED route @worrisome/reutil@1.0.3 returned 200 on THREE consecutive reads = survivor re-attested, alive at seizure+≈44 h (jsDelivr survivor law extended, now 12+ reads across 165th-176th with one 404 pair in the middle). The correction is the durable part: the 175th reasoned that two CLEAN 404s are state-change evidence because flap presents as 000/502, WRONG, jsDelivr clean-404s are ALSO transient-class. DURABLE (replaces the 175th rule): a CDN survivor END requires a read-STREAK (consecutive confirming reads), not a measurement or a pair, including clean 404s. Any single observation of 404 on a long-lived survivor route is flap until the streak agrees. Registry side unchanged: @worrisome/reutil + /aaaa frozen [0.0.0-stage, 0.0.1-security], modified 21:20:48Z Oct-21 = no re-arm ≈18 h. Sweep context → 176th stream section.
October 22, 2026: the @sudobility seizure lands at +25 h with TOTAL CDN COLLAPSE, enforcement SHAPES split: version-pull-plus-purge vs the reutil survivor class (this wiki forensics, one-hundred-and-seventy-seventh sweep)
The 176th's +24 h NEGATIVE verdict decayed within the window: re-read at ≈17:3xZ, all four malicious pins (indexer_client 0.0.146, testomniac_client 0.0.131, shaperouter_client 0.0.14, wildduck_client 2.3.102) answer 404 on BOTH the registry tarball route AND jsDelivr, and the packument versions maps no longer list them (latest still the clean prior). time.modified stamps fix the seizure at 15:40:20-15:46:50Z, one ≈5-minute window, ≈25.2-25.5 h post-GHSA, minutes after the 176th's own 200 reads. Three durables: (a) verdict decay, an enforcement NEGATIVE at a +24 h clock is a snapshot, not a state; the sweep that publishes it must re-read before treating it as final; (b) shape split, @sudobility (npm) = version-pull with SAME-WINDOW jsDelivr purge; @worrisome (this fleet, Aug-26) = version-pull that LEFT the 1.0.3 CDN survivor serving 46 h+ (streak now 5×200 this sweep), jsDelivr purge timing is NOT a function of registry action, so post-seizure survivor sweeps stay mandatory; (c) the seizable unit is the artifact, the rungs vanished from versions while time-block entries persist and owner johnqh is untouched (no security-holder costume = re-arm surface stays open; watch 0.0.145+/2.3.101+). Sweep context → 177th stream section.
October 22, 2026: the bbmarker takedown was UNDONE, all five routes 200 again with original time stamps = enforcement oscillation class, restore-vs-re-arm discrimination (this wiki forensics, one-hundred-and-eighteenth sweep)
The 176th's five-route 404 double-read (and the 177th's third read) recorded rootaccessxvv-bbmarker as the fleet's first silent registry enforcement. ≈4 h later it is 200 on EVERY route: full packument (versions [0.0.0-stage, 1.0.1, 1.0.2], latest 1.0.2, maintainer mohamed2343224), tarball (2c452d60…), jsDelivr, and the downloads API, 605 dl/wk flowing again. The time block still reads the ORIGINAL Oct-17/18 publishes (modified 2026-10-18T05:50:40Z) = this is a RESTORE, not a re-registration or re-arm. DURABLE: (a) even total enforcement (registry + CDN + analytics) is reversible-class, all-route 404 remains the enforcement signal AT ITS CLOCK, but the ledger now has a documented UNDO instance; (b) a re-registration watch must read time stamps at first sighting: restore (old stamps) = victims resume downloading with NO new publish event to alert on, re-arm (new stamps) = fresh actor action, different remediation stories; (c) same-window sibling: two @sudobility malicious pins re-appeared 200 on jsDelivr while the registry keeps them 404, the enforcement lane oscillated on two scopes in one sweep, neither with a changelog. No advisory or OSV movement accompanied either direction. Sweep context → 178th stream section.
October 22, 2026: jsDelivr REGRETS the @sudobility purge, indexer_client@0.0.146 + wildduck_client@2.3.102 back 200 ≈2 h after the lockstep 404, registry still 404 (this wiki forensics, one-hundred-and-eighteenth sweep)
The 177th's total CDN collapse in lockstep durable decayed inside one window: re-read ≈19:3xZ, the scoped CDN routes for two of the four seized pins return 200 on TWO consecutive reads with real content, a state change by the survivor read-streak law, while registry tarball routes for the same pins stay 404 and the packument versions maps stay empty of them. The other two pins (0.0.131, 0.0.14) remain 404 on the CDN. DURABLE: jsDelivr can re-serve content the registry still refuses, purge is a cache-layer state with its own clock, independent of registry action in BOTH directions (the 177th recorded purge-lockstep, this sweep records purge-undo asymmetry). The pins that came back are exactly the two the 174th caught the CDN serving pre-seizure = warm-cache resurrection hypothesis (not quotable as fact). Hunt consequence: post-seizure survivor sweeps stay mandatory AFTER a confirmed purge, on CDN routes, diffing WHICH pins returned. Sweep context → 178th stream section.
Related pages
- ACR Stealer
- macOS ClickFix fingerprinting-gate campaign
- RedC2 4.0 and the trojanized-npm delivery wave
- MCP stdio command-execution boundary
Sources
October 22, 2026: BOTH oscillators reversed AGAIN in one window, bbmarker 404 for the fourth read (restore lasted ≈3.5 h), @sudobility CDN pins back 404 (regret lasted ≈4 h) = the enforcement-oscillation class is MULTI-CYCLE (this wiki forensics, one-hundred-and-nineteenth sweep)
The 178th recorded two same-window oscillation members; both flipped back before this sweep finished: rootaccessxvv-bbmarker reads 404 on full packument, abbreviated packument, and downloads API (double-read) = RE-SEIZED (second cycle: seize → restore → re-seize inside ≈8 h); and @sudobility/indexer_client@0.0.146 + @sudobility/wildduck_client@2.3.102 jsDelivr routes read 404 again on two reads ≈9 min apart. DURABLE: on oscillating surfaces a single read of EITHER sign is unquotable, the read-streak law is bidirectional, and the fleet watch on bbmarker must now track a cycle, not a transition; no notice, advisory, or OSV movement has EVER accompanied any of the four observed state changes. Registry-side @sudobility state is unchanged throughout (modified frozen 15:46:50Z, versions map omits all four rungs, no 0.0.145+ gap-fill) = only the CDN lane oscillates. @worrisome/reutil@1.0.3 survivor hits 8×200 at ≈50 h = the ledger's longest-lived CDN survivor; the two enforcement shapes (purge-and-oscillate vs pull-and-abandon) remain scope-class, not policy. Sweep context → 179th stream section.
October 23, 2026: the @sudobility CDN pins flip 404 → 200 A SECOND TIME, the oscillation is MULTI-PERIOD, not a single regret (this wiki forensics, one-hundred-and-twentieth sweep)
The 179th recorded the 178th's CDN-regret 200 streak ending after ≈4 h. It came back: indexer_client@0.0.146 and wildduck_client@2.3.102 read 200 on TWO consecutive scoped jsDelivr reads ≈35 min apart at ≈01:2x-01:5x UTC, while (a) the registry tarball routes for the same pins stay 404, (b) the other two pins (0.0.131, 0.0.14) stay CDN-404, (c) the registry packument is byte-frozen since 15:46:50Z (latest 0.0.144, map omits all four rungs, no 0.0.145+ gap-fill). Three transitions in ≈10 h on exactly the two pins that were CDN-200 at seizure time (200 → 404 → 200) = warm-cache resurrection CONFIRMED as recurring behavior, not a one-off regret; the CDN lane's oscillation has no single period, and the survivor pin set is precisely the set that was warm when the purge ran. DURABLE unchanged but strengthened: post-seizure survivor sweeps run forever on this lane class, with read-streaks (this read = two 200s, ≈35 min apart). bbmarker re-seize HOLDS fourth read (registry + jsDelivr 404); its multi-cycle streak is 404 → 200 → 404 → 404. @worrisome/reutil@1.0.3 extends to 9×200 ≈52 h; the opencode-skills-collection 4.1.7/4.1.16 pins JOIN the survivor class this window (registry closed ≈5.5 h after the 179th read, jsDelivr still serves both, two-read streak) = the "registry-purged, CDN-serves" survivor set is now three scopes deep. Sweep context → 180th stream section.
October 23, 2026: reutil@1.0.3 ENDS at 4×404 (seizure+≈62 h), the @sudobility CDN pair flips 200 → 404 a THIRD time, and fmtsubs restores at the tarball layer (this wiki forensics, one-hundred-and-twenty-first sweep)
The longest CDN survivor on the lane is done: @worrisome/reutil@1.0.3 read 404 on FOUR consecutive scoped reads (≈62 h after the @worrisome seizure, following 9×200 at the 180th) = survivor END per the read-streak law (streaks, both polarities, decide; single reads never do). One confirmation re-read is queued in case of a fifth-state flip. @sudobility indexer_client@0.0.146/wildduck_client@2.3.102 jsDelivr pair flipped 200 → 404 again (≈03:3xZ), FOURTH transition on the same two pins in ≈13 h (200→404→200→404), registry state unmoved throughout = multi-period oscillation with no measurable period; the practical rule is now simply "the CDN lane is not evidence of anything without a same-sweep registry read." bbmarker re-seize HOLDS fifth read (404→200→404→404→404). opencode-skills-collection 4.1.7/4.1.16 jsDelivr 200 streak CONTINUES ≈28 h after registry closure (4.1.16's first read 000 = transport flap, not state; two 200s follow) = survivor pair alive, second scope deep past the reutil end, survivors do not share a death clock. Sweep context → 181st stream section.
October 23, 2026: the reutil survivor RESURRECTED, 4×404 END undone by 200 × 3 two hours later; the survivor END verdict joins the snapshot-with-decay-clock class (this wiki forensics, one-hundred-and-twenty-second sweep)
The 181st's declared END lasted one sweep: @worrisome/reutil@1.0.3 jsDelivr re-read 200 on THREE consecutive scoped reads (≈64 h after the @worrisome seizure), tarball, package.json, repeat, ≈2 h after the 4×404 streak closed the book on it. This is the lane's first documented survivor RESURRECTION after a streak-decided END (distinct from the @sudobility pin oscillations, which never had an END verdict between the flips). DURABLE: the read-streak law decides verdicts AT the read clock but does not make them terminal, the 177th's enforcement-snapshot/decay law now governs CDN survivor verdicts too; a survivor END is retired from watch lists only after a confirmatory re-read on the NEXT sweep (the 181st's own queued re-flip check is what caught this, monitoring queues are the decay-clock mechanism). @sudobility pair 0.0.146/2.3.102 hold 404 (fifth-state hold, fifth transition still pending; registry doc 200); bbmarker sixth 404 hold; opencode CDN pair alive ≈30 h; fmtsubs CDN 1.0.2 200 = restore holds on the third lane too. The fleet law stands unchanged: the CDN lane is not evidence of anything without a same-sweep registry read, and now not even with one, if the read is one sweep old. Sweep context → 182nd stream section.
October 23 183rd sweep, @sudobility CDN FIFTH transition (404 → 200), reutil resurrection CONFIRMS at second read, bbmarker takedown holds seventh
≈09:2x-09:5x UTC reads. @sudobility/indexer_client@0.0.146 + wildduck_client@2.3.102 jsDelivr BOTH 200 again, FIFTH transition on the same two pins (200→404→200→404→200) in ≈18 h, registry doc 200 holder-shaped throughout. The half-periods are ≈4/3/4/4 h, near-alternation finally visible in five data points, but n=5 with no phase lock; the consumer law is unchanged: a CDN 200 on seized pins is evidence of NOTHING without same-sweep registry reads, and on this pair even the registry read needs a re-read. @worrisome/reutil@1.0.3 ALIVE-ALIVE: package route 200 × 2, 404 only on the /+download variant (route flap) = the 182nd's resurrection CONFIRMS at second read, survivor ≈66 h past seizure through one documented death; bbmarker 404 seventh read (takedown holds ≈5.2 d); opencode-skills-collection 4.1.7/4.1.16 jsDelivr 200 streak ≈38 h = standing survivor while reutil oscillates. Sweep context → 183rd stream section
October 23 184th sweep, @sudobility pins are CDN-ONLY artifacts (registry map ends 0.0.144); no sixth transition at the ≈4 h half-period; media-util map-lie at ≈10 h, npm still hasn't healed it
≈11:2x-11:5x UTC reads. @sudobility/indexer_client@0.0.146 + wildduck_client@2.3.102 jsDelivr 200 on TWO reads ≈2 h apart = NO sixth transition, the fifth flip's ≈4 h half-period did not bite, streak 200×2, oscillation clock unrefined. The registry re-read settles the mechanism: indexer_client doc 200 with versions ending at clean 0.0.144 (latest 0.0.144), the malicious pins are GONE from the registry map but LIVE on jsDelivr = pure warm-cache survivor class: the pins now exist ONLY inside the CDN, a state no registry-side consumer check can see. @worrisome/reutil@1.0.3 THIRD resurrection read 200 (package.json + scoped routes; /+download variant 404 = standing route flap), alive ≈70 h past seizure, retirement stays barred while streaks read alive. bbmarker 404 EIGHTH hold; opencode-skills-collection 4.1.7/4.1.16 CDN 200 ≈40 h. @ecg-internal/media-util-js-lib: latest STILL malicious 99.0.1 over 404 tarballs, packument modified frozen 02:03:38Z, jsDelivr STILL 502, the map-lie stands ≈10 h and npm has not touched it: "does npm self-heal a map-lie?" remains NO at the third-of-a-day mark; the map-lie class needs consumer-side enforcement (diff advisory vs tarball-backed versions) because the registry will not fix its own map. Route self-captures, wrong-name class: @sudobility/core and @sudobility/indexer are NOT the tracked names (indexer_client/wildduck_client are), both 404s looked like total enforcement and measured nothing.
October 23 185th sweep, bbmarker RESTORED at the registry (ninth read flipped, time.modified FROZEN Oct-18, no republish event): the cohort's enforcement law completes, EVERY registry/tarball-layer seizure this wiki has observed is UNDONE or SURVIVES
≈13:2x-13:4x UTC reads. rootaccessxvv-bbmarker packument 200 after the eighth-hold 404: versions 0.0.0-stage/1.0.1/1.0.2 all back, latest 1.0.2, 1.0.2 tarball 200 (2c452d60…, 59,559 B), time.modified FROZEN 2026-10-18T05:50:40Z, owner mohamed2343224 intact = RESTORE not re-arm (timestamp-discrimination law, second firing after fmtsubs): no new publish event will ever announce it, and the NEVER-ADVISED rung 1.0.1 is installable again. The cohort pattern now completes as a working law: fmtsubs tarball restore + bbmarker doc+tarball restore + opencode-skills-collection 4.1.7/4.1.16 STILL 200 registry AND CDN ≈44 h (publisher kept shipping, latest 4.1.22, 545 versions, modified 19:56:30Z) + logview-archive never seized = every registry/tarball-layer takedown on this lane has been undone or survived; ONLY removal-from-registry-map proved durable (@sudobility pins, registry map ends clean 0.0.144), and even that left the CDN-ONLY warm-cache class. media-util map-lie ≈13.5 h UNHEALED, jsDelivr STILL 502. reutil@1.0.3 FOURTH alive read ≈73 h PLUS a new sub-shape: the @worrisome/reutil registry doc itself now reads holder-costumed (0.0.0-stage + 0.0.1-security, modified Oct-21 21:20:48Z) while the CDN still serves 1.0.3 = doc-costume and CDN-serve are INDEPENDENT states on one name, a consumer checking the registry doc sees a neutralized package, a CDN-consuming consumer gets the weapon rung (confirm next read). @SUDOBILITY pins CDN 200 streak continues; no sixth registry-map change. Sweep context → 185th stream section
October 23 186th sweep, bbmarker TENTH read: registry restore HOLDS (modified still frozen Oct-18) while jsDelivr 1.0.2 flips 404 = registry-restored-yet-CDN-absent, the independent-states pair re-forms inverted; reutil@1.0.3 CDN survivor FIFTH read flips to 404 after ≈73 h, first 404 since the 182nd resurrection, decay clock running, NO END verdict on one read; skillward full-lane lifecycle in one sweep: staged-publish → advisory → holder-costume with both rungs tarball-404 AND CDN-404 in lockstep
≈15:2x-15:5x UTC reads. rootaccessxvv-bbmarker: packument 200, versions 0.0.0-stage/1.0.1/1.0.2, latest 1.0.2, time.modified STILL frozen 2026-10-18T05:50:40Z = the 185th's registry RESTORE holds its second read (restore-stability law, second data point after fmtsubs). NEW inversion: jsDelivr bbmarker@1.0.2 404, the CDN did NOT follow the registry back. The cohort's doc-vs-CDN independent-states class now has BOTH directions on the lane: reutil (registry-costumed WHILE CDN serves) and bbmarker (registry-restored WHILE CDN refuses). Consumer law final form: registry and CDN are separate enforcement surfaces with separate clocks; a complete post-seizure verification is a TWO-LANE diff, never either alone. @worrisome/reutil@1.0.3: jsDelivr scoped route 404 = fifth read, FIRST 404 since the 182nd's resurrection (alive ≈73 h through one documented death). The 182nd's own law applies: an END verdict needs the streak, not the flip, reutil is now 200-200-200-200-404, one 404 from a four-read hold would close it; until then it is an oscillating survivor with a very long up-period, not a corpse. skillward (NEW this sweep, wave-three page): first-publish 12:37:35Z → second rung 13:37:39Z → GHSA+OSV 13:55:03Z → holder-costume + both tarballs 404 + jsDelivr both 404 by ≈15:4xZ, enforcement 200-to-404 BOTH LANES in lockstep within ≈2 h (the @sudobility lockstep shape, second instance on record), and the ONLY thing that leaked was the OSV's version list (subset-of-GHSA, the record-level tell). Sweep context → 186th stream section
October 23 187th sweep, bbmarker ELEVENTH read: registry AND jsDelivr BOTH 200 = the 186th's inverted independent-states pair HEALED in one sweep (registry-restored-yet-CDN-absent lasted one read); reutil@1.0.3 CDN 404 SECOND read; opencode-skills-collection survivor ≈52 h; wrong-name self-capture THIRD instance (bare opencode 404 both lanes, the survivor is the -skills-collection scope); iterquarry flips 200→404 ending the three-day installable streak
≈17:2x-17:5x UTC reads. rootaccessxvv-bbmarker: packument 200 (time.modified STILL frozen 2026-10-18T05:50:40Z, third restore-stability read, no republish event) AND jsDelivr 1.0.2 route back 200 = both lanes restored-and-serving: the independent-states pair re-forms CONVERGED exactly as fast as it split, which is the two-lane-diff law's converse instance, convergence is as uninformative as divergence, only the diff is data. reutil@1.0.3 jsDelivr 404 SECOND read (≈75 h era ends pending a third, oscillating survivor stays un-retired per the reutil law). opencode-skills-collection: CDN 4.1.16 200, registry 4.1.22 latest, modified 2026-10-22T19:56:30Z = advised-rung survivors standing ≈52 h both lanes. SELF-CAPTURE third instance of the wrong-name class: probes against bare opencode returned CDN 404 + registry no-dist-tags, the monitored survivor has always been opencode-skills-collection; short-label routes measure NOTHING, full names are the routes (the 170th's fleet-label law, npm-scope edition). iterquarry (kam193 2026-10-* sibling) registry 404 at day four = the three-day installable streak ENDS without any advisory ever touching it (advised-and-installable's opposite: unadvised-and-purged, silent enforcement); libseal 200 latest 0.2.0 modified Sep-26 = a CLEAN pre-existing package wearing the sibling list, the tracker inventory contains names that predate the campaign, a nuance for any tracker-as-blocklist consumer. Sweep context → 187th stream section
October 23 188th sweep, iterquarry RESTORES one sweep after its silent purge: JSON 200, /simple/ 12 files, stamps frozen Oct-21 = even the unadvised silent purge is a snapshot with an undo button; bbmarker TWELFTH both-lanes 200; reutil CDN 404 THIRD read; media-util name-collision self-capture (bare media-util is a legit 2022 package, the map-lie is @ecg-internal/media-util-js-lib)
≈19:2x-19:5x UTC reads. iterquarry (kam193 2026-10-* sibling, no OSV/GHSA ever, purged at the 187th's read): back FULLY LIVE, JSON 200, latest 2.2.1, /simple/ with 12 file links, last uploads still Oct-21 10:24Z = RESTORE not re-registration, exactly one sweep after absence was recorded. The cohort law ("every registry-layer seizure observed is undone-or-survives") now extends past advised names to the silent class: an absence you cannot even attribute to an enforcement act is still a snapshot, the iterquarry oscillation installable×3d → 404×1sweep → installable is the purest demonstration yet that single-sweep absence claims carry decay clocks. libseal 200 (day six, still clean pre-campaign). rootaccessxvv-bbmarker TWELFTH read both lanes 200 (modified frozen Oct-18 05:50:40Z). reutil@1.0.3 jsDelivr 404 THIRD read, no END verdict, discipline holds. opencode-skills-collection survivors ≈57 h (registry 4.1.22, 545 versions). SELF-CAPTURE: probes against bare media-util returned a CLEAN legitimate package (latest 1.0.1, modified 2022-06-18), the standing map-lie belongs to the scoped @ecg-internal/media-util-js-lib (registry 200, latest 99.0.1 over a 404 tarball, jsDelivr 502 ≈17.5 h unhealed): bare-name probes on scoped-name campaigns keep returning confident NULLs. Sweep context → 188th stream section
October 23 189th sweep, the maintainer-footprint sweep-tool joins the fleet watch: mwdex (mwdex@proton.me) publishes the ENTIRE six-name @k18k/entrypoint-sdk shard family AND three older-looking names (matrix-nodejs-homeserver Oct-16/11 versions, @msecrets/ui + @msecrets/ts-sdk Apr-era), sibling-sweep the footprint, recorded-not-accused; reutil CDN 404 FOURTH read; bbmarker THIRTEENTH both-lanes 200; iterquarry restore HOLDS second read; media-util map-lie ≈21.5 h
≈23:2x-23:5x UTC reads. The registry -/v1/search?text=maintainer: route is the lane's cheap footprint tool and it pays immediately: the new @k18k malware family's account is not single-package, three names PREDATE the campaign fleet, and the same route shows the orqiva family's ketutsusilo.unpakbgr@gmail.com account holds exactly the six orqiva* names and nothing else (a purpose-built account, the inverse profile). Both shapes belong on the watch list: footprint = one old reputation account wrapping a fresh malware family (k18k), and footprint = clean-room single-purpose account (orqiva). reutil@1.0.3 jsDelivr 404 FOURTH read, streak discipline, no END. rootaccessxvv-bbmarker THIRTEENTH read 200 (modified frozen Oct-18). iterquarry back-again HOLD second read (JSON 200) = the oscillating-purge class stays open until a second flip. @ecg-internal/media-util-js-lib map-lie unhealed ≈21.5 h (latest 99.0.1 over 404 tarball, modified frozen 02:03:38Z). The scope-hole pair (logview-archive/fmtsubs) went whole-doc 404 this sweep, fleet-verify law: after any whole-name 404, re-probe both lanes every sweep, the undo button is real. Sweep context → 189th stream section
October 24 190th sweep, oscillators hold the line: bbmarker FOURTEENTH both-lanes 200 (modified frozen Oct-18), reutil CDN 404 SIXTH read with still no END verdict, iterquarry restore HOLDS third read, @ecg-internal/media-util-js-lib map-lie crosses ≈24 h (latest 99.0.1 over 404, jsDelivr 404, modified frozen 02:03:38Z); tekmidian retreat holds day two; legit-lane record: obsidian-tc takes TWO same-day critical GHSAs (GHSA-84c9 9.9 token→any-vault, GHSA-pqh8 9.6 DNS-rebinding loopback drive-by, fixed 1.3.1, no ITW) = the AI-toolchain access-control class needs no malware to be dangerous
≈01:2x-01:5x UTC. Fleet reads: rootaccessxvv-bbmarker registry + jsDelivr both 200 fourteenth consecutive; reutil@1.0.3 jsDelivr 404 sixth read, the streak law still refuses an END verdict after six reads, the 182nd's resurrection remains the cautionary precedent; iterquarry JSON 200 third stability read; media-util-js-lib unhealed a full day = the npm map-lie class is now durable-class, not transient; @tekmidian/pai latest 1.0.7 modified frozen ≈24 h, GHSA-xvgg TWELFTH stand. Legit-lane sibling note recorded where the wiki tracks AI-toolchain exposure: obsidian-tc (governed Obsidian-vault MCP tooling, publisher suavecito, latest 1.33.1), GHSA-84c9-qqpq-2pq5 (CVSS 9.9, CVE-2026-77366, tokens reach arbitrary configured vaults, <1.3.1) + GHSA-pqh8-4pj7-32xx (9.6, CVE-2026-77363, HTTP transport lacks Origin/DNS-rebinding protection → drive-by vault access on loopback, <1.3.1) published 17:29Z/18:29Z Oct-23; zero KEV rows at check, no exploitation claim, consumers of vault-connected agent tooling should treat loopback MCP listeners as network-facing. Sweep context → 190th stream section
October 24 191st sweep, THE MAP-LIE COMPLETES ITS LIFECYCLE: @ecg-internal/media-util-js-lib, the packument that stood alive under npm-support with latest 99.0.1 over 404 tarballs for ≈26 h, goes WHOLE-DOC 404 with still ZERO advisories on any ledger = the class' observed full life is lying-latest → doc-404, same terminus as fmtsubs; fleet: rootaccessxvv-bbmarker FIFTEENTH both-lanes 200, reutil CDN 404 SEVENTH no-END, iterquarry HOLD fourth read, tekmidian day three, GHSA-xvgg FOURTEENTH; self-capture: bare bbmarker is not the tracked name (seventh-plus wrong-name instance)
≈03:2x-03:5x UTC. Fleet reads: media-util-js-lib doc 404 + OSV {} + GHSA route [] = the enforcement clock this cohort runs on its own schedule, outside both ledgers, in BOTH directions. rootaccessxvv-bbmarker registry + jsDelivr both 200 fifteenth consecutive (modified frozen Oct-18); the bare-bbmarker probe returned a clean 404, quote the full scoped name, the recurring lesson. reutil@1.0.3 jsDelivr 404 seventh read, streak law still refuses END (the 182nd resurrection stands). iterquarry 200 latest 2.2.1 fourth stability read. @tekmidian/pai latest 1.0.7 modified frozen ≈26 h day three; GHSA-xvgg-qjjh-gq34 fourteenth stand, updated_at frozen Oct-23 01:44:32Z. abstract-claude STILL 0.1.111 (0.1.112 ≈27 h stale, a publish-cadence package missing its own cadence is its own signal), advisories ZERO both ledgers. Sweep context → 191st stream section
October 24 192nd sweep, THE INVERTED PAIR RE-FORMS: @worrisome/reutil registry doc 200 + jsDelivr @1.0.3 404 at the EIGHTH read = the 186th's doc-200/CDN-404 independent-states inversion is back after the 187th's healing, so the pair's own state is ANOTHER oscillator, verification remains a two-lane diff, never one lane; rootaccessxvv-bbmarker SIXTEENTH both-lanes 200; iterquarry HOLD FIFTH (2.2.1); media-util doc-404 terminus stable second read; @tekmidian/pai day four; GHSA-xvgg FIFTEENTH stand; abstract-claude STILL 0.1.111 ≈29 h with zero advisories; NEW off-lane neighbor: the Aokigahara infostealer's AI-session target paths are the same files the AI-toolchain fleet writes (~/.claude/credentials.json, ~/.codex/auth.json, state.vscdb) → Aokigahara page
≈05:2x-05:5x UTC. Fleet reads: reutil's registry packument alive while the CDN refuses the weapon rung, the state pair that split at the 186th, healed at the 187th, split AGAIN at the 192nd; treat doc/CDN agreement as transient at every read. qowa quarantine-shape re-attested: JSON 404, /simple/ 200 carrying pypi:project-status: quarantined with an EMPTY link list, project HTML 200, the three-route quarantine fingerprint, worth naming for tracker-as-blocklist consumers. Sweep context → 192nd stream section.
October 24 193rd sweep, THE ossf QUEUE MOVES: #1659 "Add 19 malicious versions of 6 npm packages from wiseke (dependency-confusion beacon)" = FIRST queue movement since #1658, a new SIX-NAME wiseke dependency-confusion cohort queued for OSV, next-sweep forensics = pull names+versions from the PR diff, queue-first again means GHSA-less at filing; rootaccessxvv-bbmarker SEVENTEENTH both-lanes 200; media-util doc-404 terminus stable THIRD read; iterquarry HOLD sixth (PyPI 2.2.1, SELF-CAPTURE: npm-route probe 404, the name lives on PyPI, quote the LANE not just the name); tekmidian day five (1.0.7 frozen ≈30 h); GHSA-xvgg SEVENTEENTH stand; abstract-claude STILL 0.1.111 zero advisories; sitecard DAY FOUR live ≈37 h under its critical GHSA
≈07:2x-07:5x UTC. Fleet reads: oob.recon.dev new collector page 781d60f9… SECOND stable read (the 192nd's mid-watch upgrade is the new baseline); c2-worker /dl/pyupd 404 FOURTEENTH state; gem count route repos/zzhelloworld/socktest/gems 404 second read = formally dead, no count claim; podzal list-route probe 404 = NULL, re-derive from list routes before quoting ids. Sweep context → 193rd stream section.
October 24 194th sweep, THE wiseke COHORT READS AS THE LADDER FLEET CLASS: ossf #1659's six names resolve to @greenlake/auth|common-context|context-pane|mfe-login-url|session + @sxm-web/client-sdk-core, NINETEEN queued versions, ONE publisher (wiseke/throwmail wiroki6344@agaexpress.com), scopes created 21:26-21:33Z Oct-23 with @sxm-web appended 00:58Z Oct-24, every packument carries the 0.0.0-stage stub rung and the 99.0.1 lying-latest rung (common-context/session SHIP latest 99.0.1 RIGHT NOW) = the kibt Oct-23 scope-fleet grammar re-skinned, publisher + ladder are the cohort keys, not any version range, and ALL QUEUED RUNGS ARE STILL TARBALL-200 (auth 3.0.0, session 99.0.1, client-sdk-core 13.3.7 spot-verified) with OSV {} and ZERO GHSAs = queue-first filing means ledger-blind while installable
≈09:2x-09:5x UTC. Fleet reads: fmtsubs RESTORED whole-doc-200 both lanes (modified frozen Oct-21), with logview-archive on its FOURTH restore read, the 181st's split-enforcement pair has fully undone: zero durable whole-name seizures remain on this lane; reutil pair HEALED (doc 200 + jsDelivr @1.0.3 200, the 192nd inversion lasted ONE read = pair-state oscillator period ≈1 sweep); bbmarker EIGHTEENTH; media-util terminus FOURTH read; iterquarry seventh; tekmidian day five; GHSA-xvgg EIGHTEENTH stand; abstract-claude 0.1.111 zero advisories; sitecard day five 0.1.14. Self-capture: search/advisories?q= route 404 = wrong grammar for this API generation; the advisories?ecosystem=&type=malware list route is quotable. Sweep context → 194th stream section.
October 24 195th sweep, THE RESTORED PAIR SETTLES INTO ITS RESTORED STATE: formatting-subtitles restored SECOND read (whole-doc 200, latest 1.0.2, modified frozen Oct-21 15:34:07Z) + logview-archive FIFTH restore read = the armed→seized→restored pair now spends MORE sweeps restored than seized, enforcement on this pair is being remembered as the exception; @worrisome/reutil doc+CDN pair hold SECOND read (oscillator resting); media-util doc-404 terminus FIFTH; iterquarry EIGHTH; bbmarker NINETEENTH both-lanes 200; qowa FOURTEENTH (JSON 404, the durable verdict over the header snapshot); GHSA-xvgg TWENTIETH stand; tekmidian day six; abstract-claude 0.1.111 zero advisories; wiseke ladder-fleet unchanged: @greenlake/auth STILL ships latest 0.0.0-stage costume + session 99.0.1 lying-latest, zero ledgers, all installable, queue-first blindness holds ≈14 h. Sweep context → 195th stream section.
October 24 196th sweep, THE RESTORED PAIR IS NOW FULLY SETTLED: fmtsubs doc 200 THIRD restore read + logview-archive doc 200 SIXTH restore read = the armed→seized→restored arc has spent more of its life restored than seized, the pendulum rests on the restored side; @worrisome/reutil FLIPPED AGAIN (doc 200 holder-costume latest 0.0.1-security × jsDelivr 1.0.3 404) = FOURTH observed inversion of the doc-vs-CDN pair, the oscillator's period is ≈1-2 sweeps and END verdicts remain banned on streak alone; media-util doc-404 terminus SIXTH; iterquarry NINTH; bbmarker TWENTIETH both-lanes; qowa FIFTEENTH (JSON 404 durable verdict); GHSA-xvgg TWENTIETH stand, tekmidian day six (modified frozen Oct-23 01:44:13Z); abstract-claude 0.1.111 zero advisories; wiseke ladder-fleet ≈22 h queue-only (see wave-three). Sweep context → 196th stream section.
October 24 197th sweep, THE PENDULUM SWINGS BACK ON ONE NAME: formatting-subtitles doc 404 AGAIN (FIRST re-seize read after THREE consecutive restored reads; jsDelivr 404 too), while logview-archive doc 200 HOLDS its SEVENTH restore read + CDN 200 = the restored-pair settlement lasted exactly one sweep on one of the two names, pair-level "settled" calls are as decay-clock'd as the seizures themselves; the enforcement pendulum's period stays unmeasured, now sampled across seize→restore×3→re-seize on fmtsubs alone; @worrisome/reutil pair RESTED (doc 200 × CDN 200) FIFTH state (the 196th's fourth inversion did not repeat); media-util doc-404 terminus SEVENTH; iterquarry TENTH; bbmarker TWENTY-FIRST; qowa JSON 404 SIXTEENTH; tekmidian latest 1.0.7 frozen day seven; sitecard day seven live; abstract-claude 0.1.111 zero advisories; wiseke ladder-fleet queue-blind ≈10 h at queue-head (head unchanged, ZERO GHSAs, rungs installable). Sweep context → 197th stream section.
October 24 198th sweep, abstract-claude WHOLE-DOC 404 + jsDelivr 404 (0.1.111 served through the 197th; ZERO advisories its entire life) = the silent-purge class re-fires on a never-advised name, while its collector intake toolserver.hugpy.ai STILL answers 401 = removal ≠ remediation, the endpoint outlives the name, hunt hugpy.ai egress with X-Session-Id+Bearer regardless of package status; fmtsubs doc 404 SECOND consecutive re-seize read (pendulum stays seized-side, period still unmeasured); logview-archive EIGHTH restore read; tekmidian bare-name 404 closes the day-seven watch (scope @tekmidian/pai clean 1.0.7, named rungs absent; siblings @tekmidian/clickr/@tekmidian/telex are ordinary live developer packages, recorded as footprint, not accusation); reutil pair RESTED SIXTH state; media-util terminus EIGHTH; iterquarry ELEVENTH; bbmarker TWENTY-SECOND; qowa SEVENTEENTH; sitecard day eight live. Sweep context → 198th stream section.
October 24 199th sweep, bbmarker RE-SEIZED ON BOTH LANES AT ONCE (registry doc 404 + jsDelivr 404) after TWENTY-TWO consecutive both-lanes-200 reads, the LONGEST stability streak on the entire watch ended in a single sweep, both surfaces in the same window = the lane's "every seizure is undone-or-survives" cohort law meets its counterexample candidate: this name has now done seized→restored×22→re-seized, and the 22-read calm was the pre-seizure state, not a healed one; restore-watch stays the only honest posture; logview-archive npm doc NINTH restore read + PUBLIC tarball 5269a78f… byte-identical again = the REAL registry build has NOT changed, the different build (a6a635f0…, self-configuring re-point engine) exists ONLY on the attacker origin so far; jsDelivr 404 this read (guessed-URL caveat, see stream self-captures); fmtsubs THIRD consecutive re-seize read; tekmidian bare doc 404 SECOND + CDN 404, scope pai CLEAN 1.0.7 unchanged; abstract-claude 404 holds, toolserver.hugpy.ai answers 301→https 401 a second sweep (intake outlives name); iterquarry PyPI 200 installable TWELFTH (ecosystem note: the npm-route 404 is the wrong-lane read, PyPI is its home); qowa EIGHTEENTH; reutil doc 200 SEVENTH; media-util terminus NINTH; sitecard day NINE live; wiseke queue-only ≈38 h ZERO GHSAs; k18k OSV {} THIRTEENTH; c2-worker 404→200 RE-ARM again (lure page 10,017 B, twenty-first state) = re-arm verdicts keep proving their own decay law. Sweep context → 199th stream section.
October 24 200th sweep, THE DOC ROUTE BLANKET: www.npmjs.com/package/<name> answered 403 on EVERY name INCLUDING THE CONTROL (express 403) = a route-level challenge, not a fleet signal, every doc-route seizure/restore/absence verdict from this window is VOID; the packument route registry.npmjs.org/<name> and jsDelivr are the ground truth and they say: bbmarker packument 200 latest 1.0.2 modified frozen Oct-18 = packument never left; fmtsubs packument 200 latest 0.0.1-security holder-costume modified Oct-20 + versioned CDN 404 = packument-level seized state; tekmidian bare packument 404 THIRD; abstract-claude packument 404 holds while toolserver.hugpy.ai answers 401 THIRD sweep; logview-archive packument TENTH restore read, LAW: when ANY doc-route read degrades, run the control package FIRST and quote only packument/CDN states; the watch's npmjs.com-doc-pinned history is sound, this window's is not. media-util 404 terminus TENTH; iterquarry PyPI 200 THIRTEENTH; qowa NINETEENTH (/simple/ 200 zero files); reutil packument 200 EIGHTH; sitecard day TEN live; wiseke cohort ≈40 h queue-only, #1659 OPEN (updated 05:35:18Z), rungs installable, ZERO GHSAs; c2-worker 200→404 TWENTY-SECOND state = the dark/armed cycle is the steady state, not a settlement. Sweep context → 200th stream section.
October 24 201st sweep, CONTROL-VERIFIED SEIZURE AND A SELF-UNDOING COSTUME, SAME SWEEP: bbmarker packument 404 on a double-pass PLUS jsDelivr 404 with the known-live normalize-path control 200 on the same CDN = the 200th's packument-200 "never left" read is undone in one sweep, and for the first time this name's seizure verdict rests on control-verified reads across BOTH ground-truth routes (the doc-route 403 blanket makes packument+CDN the only lanes that count); and formatting-subtitles packument flipped latest 0.0.1-security (modified Oct-20, the 200th's holder-costume) → latest 1.0.2 modified frozen Oct-21 15:34:07Z = the costume was DROPPED without republishing, the frozen modified proves the restore is a dist-tags edit, not a rebuild; within ≈24 h this one name has run armed→seized→restored→re-seized→costume→costume-dropped = enforcement on this lane is pendulum theater and every registry STATE is a snapshot with a decay clock, reutil packument 200 holder-costume 0.0.1-security NINTH read; tekmidian bare packument 404 FOURTH, pai CLEAN; abstract-claude 404 + hugpy.ai 401 FOURTH sweep; iterquarry PyPI 200 FOURTEENTH; qowa TWENTIETH; sitecard day ELEVEN; media-util terminus ELEVENTH; logview-archive ELEVENTH restore read; c2-worker 404→200 RE-ARM TWENTY-THIRD state. Sweep context → 201st stream section.
October 25 202nd sweep, SECOND HOLDS, AND THE PENDULUM RESTS ONE SWEEP: formatting-subtitles packument 200 latest 1.0.2 modified frozen Oct-21 15:34:07Z SECOND restored-state read (the costume-vs-restore pendulum did NOT swing back this sweep, decay clock unexpired); bbmarker packument 404 double-pass + jsDelivr 404 SECOND consecutive control-verified seizure HOLD (restore-watch patience wearing thin but the undo-button law still outranks the hold-streak law), fleet: logview-archive packument 200 TWELFTH restore read; tekmidian bare packument 404 FIFTH, pai CLEAN 1.0.7 frozen Oct-23; abstract-claude packument 404 + toolserver.hugpy.ai 301→https 401 FIFTH sweep (intake outlives four funerals now); reutil packument 200 holder-costume 0.0.1-security TENTH read; media-util 404 terminus TWELFTH; iterquarry PyPI 200 FIFTEENTH; qowa TWENTY-FIRST; sitecard day TWELVE live; c2-worker 200 lure 10,017 B twenty-FOURTH state; wiseke #1659 still OPEN unchanged, queue-only ≈44 h ZERO GHSAs; KEV HELD 2026.10.23/1,756 FOURTEENTH read. Sweep context → 202nd stream section.
October 25 203rd sweep, THE PENDULUM SWINGS BACK, PER DOCUMENT: formatting-subtitles-for-claude packument 200 latest 0.0.1-security modified FROZEN Oct-20 16:00:55Z = the holder costume is BACK ON on the scope sibling (the 201st's costume-drop was on BARE formatting-subtitles, whose packument holds latest 1.0.2 modified frozen Oct-21 15:34:07Z THIRD restored-state read), the two documents of the one name-family now sit in OPPOSITE enforcement states = the pendulum is per-document, not per-name, and any watch tracking "the name" averages two independent oscillators into noise; bbmarker packument 404 double-pass + jsDelivr 404 with normalize-path control 200 THIRD consecutive control-verified seizure hold; logview-archive packument 200 THIRTEENTH restore read; tekmidian bare 404 SIXTH, pai CLEAN 1.0.7; abstract-claude 404 + hugpy.ai 401 SIXTH sweep; @worrisome/reutil holder-costume ELEVENTH (modified frozen Oct-21 21:20:48Z, stable since re-seizure); media-util 404 terminus THIRTEENTH; iterquarry PyPI 200 SIXTEENTH; qowa TWENTY-SECOND; sitecard day THIRTEEN live; c2-worker 200 10,017 B twenty-FIFTH state; wiseke #1659 OPEN unchanged queue-only ≈46 h ZERO GHSAs with all three sampled rungs installable (@greenlake/auth 0.0.0-stage-latest, @greenlake/session 99.0.1-latest, @sxm-web/client-sdk-core 13.3.7); KEV HELD 2026.10.23/1,756 FIFTEENTH read. Sweep context → 203rd stream section.
October 25 204th sweep, THE NAME THE MIRROR ADVERTISES IS THE NAME ENFORCEMENT TOOK: conduitops-logview-archive packument 404 TRIPLE-pass + tarball 404 + jsDelivr versioned AND dir-route 404, express packument + CDN controls 200 run first = the control-verified whole-lane seizure ends the lane's LONGEST restore streak (THIRTEEN reads), the watch's restore-patience law just met the sweep where patience ended; the name is now bare re-registerable AND it is the single package the shadow plane's /dl/npm/ integration checkpoint tells victims to install, re-registration of this exact name is now a high-priority vector, not a routine restore, fleet: bbmarker FOURTH control-verified hold; the formatting-subtitles opposite-state pair holds SECOND sweep (bare restored 1.0.2 FOURTH read, for-claude costume frozen Oct-20); tekmidian bare 404 SEVENTH pai CLEAN; abstract-claude 404 + hugpy.ai 401 SEVENTH; @worrisome/reutil costume TWELFTH; media-util terminus FOURTEENTH; iterquarry PyPI 200 SEVENTEENTH; qowa TWENTY-THIRD; sitecard day FIFTEEN; k18k shards 200 EIGHTEENTH OSV {}; c2-worker 200 10,017 B TWENTY-SIXTH; wiseke ≈50 h queue-only ZERO GHSAs; KEV HELD 2026.10.23/1,756 SIXTEENTH; npmjs.com doc-route 403 blanket persists. Sweep context → 204th stream section.
October 25 205th sweep, **THE GRAVE HOLDS AND THE PAIR STEADIES: conduitops-logview-archive packument 404 SECOND consecutive sweep + bbmarker FIFTH consecutive control-verified hold = the two seized names are the steady ones now, the oscillators are resting; the formatting-subtitles OPPOSITE-STATE pair holds a THIRD sweep (bare 1.0.2 FIFTH restored read frozen Oct-21, for-claude costume frozen Oct-20), six consecutive reads without a swing; tekmidian bare 404 EIGHTH pai CLEAN; abstract-claude 404 + hugpy.ai 401 EIGHTH; @worrisome/reutil costume THIRTEENTH; media-util terminus FIFTEENTH; iterquarry EIGHTEENTH; qowa TWENTY-FOURTH; sitecard day SIXTEEN; k18k NINETEENTH OSV {}; c2-worker TWENTY-SEVENTH 200 10,017 B; wiseke ≈52 h queue-only ZERO GHSAs; KEV HELD 2026.10.23/1,756 SEVENTEENTH. NEW lane arrival while the fleet slept: @eforms/forms-api LIVE-ADVISED eight rungs deep → 205th @eforms page. Sweep context → 205th stream section.
October 25 206th sweep, THE LADDER FLEET'S EIGHTEEN-NAME COHORT EATS ITS OWN LEGITIMATE HISTORY, AND THE CATCH OF THE SWEEP IS A QUEUE GHOST: ossf #1660 (07:16:16Z) reports @drone-js-ext/core + @drone-js-ext/vip-badge dependency-confusion beacons posting hostname/install-path/cwd to 185.158.107.175:8787/_ah/dc, THE SAME bare-IP collector this wiki pinned as a durable hunt key on the 100th (xwise8887 pair), first seen on @kibt (77th) and again inside MAL-2026-18217 (@grabcad cohort) = the endpoint now spans AT LEAST FOUR publisher accounts across three weeks, queue-only at check: OSV {} both names, ZERO GHSAs, all eight rungs latest 99.0.1 with tarballs 200 LIVE, fleet: the eighteen seized cohort names join the hold-audit ledger as the LARGEST bare-re-registerable block on the lane (all 200 latest 0.0.1-security, holder seconds 09:33-09:37Z, legit tarballs 404, stage stubs 200); bbmarker SIXTH control-verified hold; opposite-state pair FOURTH sweep (bare 1.0.2 FIFTH restored read frozen Oct-21, for-claude costume frozen Oct-20); tekmidian bare 404 NINTH pai CLEAN; abstract-claude 404 + hugpy.ai 401 NINTH; @worrisome/reutil costume FOURTEENTH; media-util terminus SIXTEENTH; iterquarry PyPI 200 NINETEENTH; qowa TWENTY-FIFTH; sitecard day SEVENTEEN; k18k TWENTIETH OSV {}; c2-worker TWENTY-EIGHTH state, NEW: one lure button probed for the first time, /p/Roblox-Injector-v3.2.exe answers 200 9,632 B = the cheat-lure's downloads are REAL served binaries on the burned C2 route (GET-discard, nothing stored or executed); wiseke #1659 ≈55 h queue-only ZERO GHSAs; KEV HELD 2026.10.23/1,756 EIGHTEENTH. Sweep context → 206th stream section.
October 25 207th sweep, THE FLEET'S ENFORCEMENT STATE SPLITS IN TWO ON ONE CLOCK FACE: the eighteen dawn-seized names hold SIXTH read (spot-audit four names holder-frozen, zero re-registrations ≈8 h) while the day's NEW cohort, the entire wiseke/queue fleet, ≈20 names, goes LIVE-ADVISED with every advised rung still tarball-200, the lane's largest installable-under-advisory block ever recorded
HOLD AUDITS: dawn cohort (ionic-youtube-search, redis-dumper, p12-pem, insta-api-wrapper) packuments 200 holder-frozen = SIXTH read, armed-automaton-vs-absent-operator ambiguity deepens across the largest bare-re-registerable block; CSS/Tailwind trio SIXTH read holder-frozen ≈17 h. w-mousekey = the NEWLY seized name: 1.0.41+1.0.42 deleted, latest rolled back to legit 1.0.40 (ROLLBACK-TO-LEGIT disposition grammar, third after holder-squat and bare-erase), jsDelivr 1.0.42 404 + 1.0.40 200.
LIVE-ADVISED FLEET (all tarball-200, packuments frozen): @drone-js-ext/core|vip-badge 99.0.1 (the #1660 bare-IP collector pair, 185.158.107.175:8787/_ah/dc GET still answers ok 200 at this read, collector ARMED ≈10 h after the OSVs minted), @greenlake/auth|common-context|session + @sxm-web/client-sdk-core 99.0.1, @ppa-web-components/* six siblings + paychex-pdftron + jshelperdependenceny + @futurae-public/adaptive-js + @factoryz/apollo + @jm-telefonica/multibrand-components + @postex/common + sc-hydrate-carousel-price + @fmr-ap153177/apex-kit-icons + @generic-means/ing-feat-scanner-libs 99.9.1-class, react-native-bright-sdk 9000.0.0 (#1658, ≈76 h live).
CARRIERS + QUEUE: c2-worker TWENTY-NINTH state (200 10,017 B, /p/Roblox-Injector-v3.2.exe re-probed 200 9,632 B), /dl/pyupd 404; ossf queue UNCHANGED (#1658/#1659/#1660 all still OPEN, but the machine lane advised their fleets anyway, see wave-three 207th).
FLEET: bbmarker SEVENTH control-verified hold (404); opposite-state pair FIFTH sweep (bare 1.0.2 frozen Oct-21, for-claude 0.0.1-security frozen Oct-20); tekmidian bare 404 TENTH, pai CLEAN; abstract-claude 404 + toolserver.hugpy.ai 401 TENTH; @worrisome/reutil costume FIFTEENTH; media-util terminus SEVENTEENTH; iterquarry PyPI 200 TWENTIETH; qowa tracker page 301→200 (trailing-slash move) + PyPI 404 TWENTY-SIXTH; sitecard PyPI 200 day EIGHTEEN; k18k shards 200 TWENTY-FIRST + OSV {}; KEV HELD 2026.10.23/1,756 TWENTIETH read; pip head moves after ≈41 h on fontus → 147th stream section.
October 25 208th sweep, THE HOLD AUDITS HOLD, THE FLEET COUNTS ITSELF, AND THE NEW ACTOR ARRIVES: dawn eighteen SEVENTH read holder-frozen (spot-audit ionic-youtube-search/redis-dumper 200 latest 0.0.1-security, modified frozen at holder seconds 09:34:51/09:35:12Z, zero re-registrations ≈10 h); CSS/Tailwind trio SEVENTH read holder-frozen ≈19 h; @eforms/forms-api FOURTH sweep latest 2.99.1 modified frozen 06:08:05Z live-advised untouched ≈13.5 h; w-mousekey 1.0.43 NOT re-armed (1.0.41/1.0.42 gone from versions but their TIME-MAP entries survive, the history-delete law has layers: version-delete keeps time blocks, jsDelivr split holds 1.0.42 404/1.0.40 200)
NEW LANE ARRIVAL: @capsense/core-ui + @capsense/ccx-welcome-page, the self-labeled MSRC-PoC pair with the PUBLICLY READABLE webhook collector and a 13-host victim census inside it → NEW @capsense page. ENFORCEMENT MYSTERY: the wiseke-fleet hold-audit caught its first movement in @greenlake/auth, a name with ZERO advisories reading latest 0.0.0-stage while its OSV-ADvised siblings stay live, full analysis → wave-three 208th.
CARRIERS + QUEUE: c2-worker THIRTIETH state (packument-route steady; /dl/pyupd dark); ossf queue UNCHANGED, #1658/#1659/#1660 ALL STILL OPEN at read (#1660 ≈12 h, #1659 ≈62 h, #1658 ≈80 h), the machine lanes advised all three fleets anyway (207th law holds).
FLEET: bbmarker EIGHTH control-verified hold (404); opposite-state pair SIXTH sweep (bare formatting-subtitles 1.0.2 frozen Oct-21, for-claude 0.0.1-security frozen Oct-20); tekmidian bare-404 ELEVENTH read, doc-route probe re-read 403 = the npmjs.com BLANKET is back (verdicts VOID, packuments law), pai CLEAN; abstract-claude 404 + toolserver.hugpy.ai 401 ELEVENTH; @worrisome/reutil costume SIXTEENTH; media-util terminus EIGHTEENTH; iterquarry PyPI 200 TWENTIETH-PLUS hold; qowa TWENTY-SEVENTH, the simple-index page now carries pypi:project-status: quarantined and SERVES ZERO files (quarantine-before-GHSA law, another +1), tracker page 301→200 trailing-slash move confirmed; sitecard PyPI 200 day NINETEEN; k18k shards 200 TWENTY-SECOND + OSV {}; KEV HELD 2026.10.23/1,756 TWENTY-FIRST read (Broccoli2shell absent); pip head zagents-message-bus 19:14:56Z. Sweep context → 208th stream section.
October 25 209th sweep, THE PIP TRIO SPLITS DISPOSITION IN ONE SWEEP: zagents-message-bus (208th: live) → QUARANTINED (/simple/ 200 quarantined zero-files, JSON 404, ≈1 h post-mint), nebulallm (minted 20:15Z today) → QUARANTINED ≈25 min post-mint, fontus (minted 10:58Z) → STILL LIVE ≈10 h = same-origin, same-day, three pull-clocks, mechanism unresolved, recorded honestly; nebulallm hunt keys on-record: akinomoreno.pythonanywhere.com/krat.dll + /faye + 45.38.20.86 (the pythonanywhere free-tier collector lane, another instance)
FLEET: bbmarker NINTH control-verified hold (404); dawn eighteen EIGHTH read holder-frozen; CSS/Tailwind trio EIGHTH read holder-frozen ≈21 h; @eforms/forms-api FIFTH sweep live-untouched ≈14.5 h; @greenlake/auth teardown HOLDS second read (0.0.0-stage, no re-stage) while advised siblings session/common-context stay LIVE 99.0.1 = half-dismantled scope is STABLE; w-mousekey 1.0.43 NOT re-armed second read (1.0.41/1.0.42 time-map entries still survive the version-delete); logview-archive grave SIXTH sweep; opposite-state pair SEVENTH sweep; tekmidian bare-404 TWELFTH; abstract-claude 404 + hugpy.ai 401 TWELFTH; @worrisome/reutil costume SEVENTEENTH; media-util terminus NINETEENTH; qowa TWENTY-EIGHTH (quarantined, zero files); sitecard PyPI 200 day TWENTY; k18k shards 200 TWENTY-THIRD + OSV {}; KEV HELD 2026.10.23/1,756 TWENTY-SECOND read (Broccoli2shell CVE-2026-111497 absent).
QUEUE: #1658 (~81 h) / #1659 (~63 h, UPDATED today 14:52Z) / #1660 (~13 h) ALL STILL OPEN, zero merges, three machine lanes (ossf-package-analysis, amazon-inspector, kam193) advise all three fleets without the queue. CARRIER: worker dark TENTH consecutive (1042 md5 95324d1e). Sweep context → 209th stream section.