Back to skill

Security audit

Windows Embedding Setup

Security checks for vulnerabilities and agentic risk

Overview

The skill has a legitimate Windows local-embedding setup purpose, but it directs users to install unpinned native npm code into the OpenClaw application directory and to ignore vulnerability warnings.

Review this skill carefully before installing. Use it only if you are comfortable modifying the OpenClaw installation and user memory configuration on Windows. Prefer a pinned, reviewed node-llama-cpp version or lockfile, do not ignore high or critical npm vulnerability reports, and verify the downloaded GGUF model with a trusted hash or immutable release before loading it.

Vulnerability Patterns
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
Findings (3)

T08 · Insecure Dependencies

Warning
Location
SKILL.md:58
Finding

Unpinned npm Dependency Installed Directly into the OpenClaw Application Directory

Content
View full analysis

Vulnerability Details

File Location: SKILL.md:58-60, SKILL.md:133-135, references/windows-setup.md:44-46
Vulnerability Type: Unpinned third-party dependency installation
Risk Level: Medium

Vulnerable Code

powershell
cd "D:\Program Files\OpenClaw\resources\openclaw"
npm install node-llama-cpp

The same installation procedure appears in both the primary Skill instructions and the Windows setup reference.

Technical Analysis

The command installs node-llama-cpp without specifying an exact version, lockfile, package integrity value, or approved release digest. npm therefore resolves the package version from the registry at installation time, meaning the installed code can differ from the version assessed when this Skill was published.

npm dependencies can execute lifecycle scripts during installation. This package can also involve native components or downloaded precompiled binaries. Installing it directly into the deployed OpenClaw application directory expands the effect of a compromised or unexpectedly changed dependency beyond an isolated development environment.

This is a supply-chain weakness rather than evidence that node-llama-cpp is malicious.

Attack Path

  1. An attacker compromises the package registry account, an upstream release, a transitive dependency, or a binary distribution endpoint.
  2. A malicious or vulnerable package version becomes the version selected by the unpinned installation command.
  3. A user follows the Skill and runs npm install node-llama-cpp.
  4. npm downloads the mutable package and dependency graph.
  5. Package lifecycle logic or native-component installation runs with the privileges of the invoking user.
  6. The altered dependency is placed in OpenClaw’s application tree and can subsequently execute when local embedding functionality is loaded.

Impact Assessment

Successful exploitation could execute code with the privileges of the acco ...[truncated 437 chars]

Remediation
View remediation

Remediation Suggestions

  1. Pin node-llama-cpp to an exact version that has been tested and approved by OpenClaw.
  2. Supply a reviewed lockfile and use npm ci rather than an unconstrained npm install.
  3. Document expected package and binary integrity hashes and verify them before use.
  4. Prefer an official OpenClaw release that bundles or explicitly declares a compatible dependency version.
  5. Review package lifecycle scripts and the provenance of native binaries before installation.
  6. Perform installation with the least-privileged account capable of updating the intended application directory.
  7. Consider installing and validating the dependency in an isolated staging copy before modifying the deployed application.

T09 · Insecure Skill Coding Practices

Warning
Location
SKILL.md:62
Finding

Instructions Explicitly Dismiss Dependency Vulnerability Reports

Content
View full analysis

Vulnerability Details

File Location: SKILL.md:62-64, references/windows-setup.md:48-50
Vulnerability Type: Unsafe handling of dependency security warnings
Risk Level: Medium

Vulnerable Instruction

The installation guidance states, in English translation:

text
The appearance of "added xxx packages" indicates success.
Warning and vulnerability messages can be ignored.

This guidance follows the installation command:

powershell
npm install node-llama-cpp

Technical Analysis

The instructions make no distinction between harmless compatibility notices and actionable security vulnerabilities. They direct users to ignore npm vulnerability output without reviewing severity, affected dependency paths, exploitability, available patches, or whether a vulnerable component is reachable through OpenClaw.

A successful installation does not establish that the resulting dependency graph is secure. Blanket suppression of audit results can cause known vulnerable transitive dependencies to remain deployed in the application.

Attack Path

  1. The dependency graph contains a package with a known exploitable vulnerability.
  2. npm reports that vulnerability during or after installation.
  3. The user follows the Skill’s instruction and disregards the report.
  4. The vulnerable dependency remains installed in OpenClaw’s application directory.
  5. OpenClaw or its local embedding subsystem invokes the affected code path.
  6. An attacker supplies the input or environmental condition required by the dependency vulnerability.
  7. The attacker obtains the impact associated with that specific dependency flaw.

The final exploitation mechanism depends on the vulnerability reported by npm; the reviewed files do not identify a particular CVE or prove that one is currently exploitable.

Impact Assessment

Impact depends on the ignored vulnerability and its reachability. Plausible consequences incl ...[truncated 245 chars]

Remediation
View remediation

Remediation Suggestions

  1. Remove the statement that warnings and vulnerabilities can be ignored.
  2. Require users to capture and review npm audit output before accepting the installation.
  3. Document which warnings are expected and explain why each one is non-security-relevant.
  4. Establish an acceptance policy based on vulnerability severity, reachability, and available fixes.
  5. Block deployment when critical or high-severity reachable vulnerabilities are present.
  6. Pin and periodically reassess the complete dependency graph.
  7. Provide a tested dependency version and known audit baseline for the supported OpenClaw release.

T08 · Insecure Dependencies

Note
Location
SKILL.md:39
Finding

Downloaded GGUF Model Is Not Cryptographically Verified

Content
View full analysis

Vulnerability Details

File Location: SKILL.md:39-52, references/windows-setup.md:8-25
Vulnerability Type: Insufficient integrity verification of an external model artifact
Risk Level: Low

Vulnerable Code and Download Reference

text
https://huggingface.co/nomic-ai/nomic-embed-text-v1.5-GGUF/resolve/main/nomic-embed-text-v1.5.Q8_0.gguf
powershell
$fs = [System.IO.File]::OpenRead("path-to-file\nomic-embed-text-v1.5.Q8_0.gguf")
$buf = New-Object byte[] 4
$fs.Read($buf, 0, 4) | Out-Null
$fs.Close()
[System.Text.Encoding]::ASCII.GetString($buf)  # Expected output: GGUF

The documentation also recommends checking that the file is approximately 139 MB.

Technical Analysis

The download URL references the mutable main branch rather than an immutable repository revision. The prescribed validation checks only the approximate file size and the four-byte GGUF magic header.

These checks establish that a file superficially resembles a GGUF artifact, but they do not authenticate its publisher or prove that its contents match the intended model. An altered file can preserve the expected size and header. No SHA-256 digest, signed manifest, immutable commit identifier, or equivalent cryptographic verification is provided.

Because the downloaded artifact is later parsed by native model-loading components, a replaced or malformed file could also expose vulnerabilities in the parser. No such parser vulnerability is demonstrated by the reviewed project.

Attack Path

  1. An attacker compromises the upstream repository, publisher account, mutable branch contents, artifact hosting path, or another part of the distribution chain.
  2. The attacker replaces the model with a modified file that retains the GGUF header and an approximately expected size.
  3. A user downloads the file from the documented mutable URL.
  4. The documented four-byte and size checks succeed.
  5. OpenClaw loads the a ...[truncated 610 chars]
Remediation
View remediation

Remediation Suggestions

  1. Replace the mutable main URL with a URL tied to an immutable repository commit or release revision.
  2. Publish the expected SHA-256 digest for the exact GGUF file.
  3. Require cryptographic verification before configuration or loading, for example:
powershell
$expected = "REPLACE_WITH_REVIEWED_SHA256"
$actual = (Get-FileHash `
  "D:\path\nomic-embed-text-v1.5.Q8_0.gguf" `
  -Algorithm SHA256).Hash.ToLower()

if ($actual -ne $expected.ToLower()) {
    throw "Model integrity verification failed."
}
  1. Obtain the digest through an authenticated channel separate from the mutable artifact when possible.
  2. Retain the GGUF header check only as a format sanity check, not as a security control.
  3. Keep the native GGUF parser updated and load untrusted models with least privilege and appropriate resource limits.
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
Findings (1)

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The skill description and scope are written as Windows-specific and explicitly target Chinese-speaking users/context without offering any language choice or opt-in. Under the policy rule, forcing a specific language/locale is a natural-language policy concern unless the constraint is clearly documented as region-specific and justified; here Windows is justified, but the Chinese-only presentation and audience framing are not offered as an option.

Content

No source excerpt is available for this finding.