Back to skill

Security audit

营销宣传册设计 Marketing Brochure

Security checks across malware telemetry and agentic risk

Overview

This skill is a disclosed dLazy cloud image-generation workflow for brochure design, with real but expected risks from running an npm CLI, storing an API key, and uploading selected media.

Before installing, review the @dlazy/cli package/source, prefer npx or an isolated environment over global install when possible, confirm the intended CLI version, and use a scoped dLazy API key that you can rotate or revoke. Only provide media paths you intend to upload to dLazy.

Vulnerability Patterns
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • 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
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
Findings (2)

T08 · Insecure Dependencies

Warning
Location
SKILL.md:48
Finding
Execution of an Unreviewed Third-Party npm Package## Vulnerability Details **File Location**: `SKILL.md:48` **Vulnerability Type**: Third-party supply-chain dependency execution **Risk Level**: Medium **Vulnerable Code Snippet**: ```bash npx @dlazy/cli@1.2.3 <command> ``` ### Technical Analysis The Skill instructs the agent to execute `@dlazy/cli@1.2.3` directly from the npm ecosystem. The implementation of that package is not included in the audited artifact, which contains only two Markdown files. Consequently, its installation scripts, transitive dependencies, runtime behavior, network activity, and handling of credentials or local files cannot be verified from this project. Running a package through `npx` may download and execute package code with the permissions of the invoking user. Although the dependency is pinned to a version, version pinning alone does not establish package integrity or protect against a compromised publisher account, malicious release, registry compromise, or unsafe transitive dependency. The Skill also supports global installation using `npm install -g @dlazy/cli@1.2.3`, which can increase the duration and system-wide scope of the installed component. The documentation contains a provenance inconsistency: one statement identifies the fixed install version as `1.0.9`, while the executable installation and `npx` instructions use `1.2.3`. This inconsistency can impede dependency review and verification. ### Attack Path 1. A user activates the brochure-generation Skill. 2. The Skill directs the agent to install or execute `@dlazy/cli@1.2.3`. 3. `npx` retrieves the package and its dependencies from the configured npm registry if they are not already cached. 4. npm package lifecycle scripts or runtime code execute with the invoking user's permissions. 5. If the package, publisher account, registry response, or transitive dependency is compromised, attacker-controlled code can run locally. 6. The malicious code could access data available t ...[truncated 945 chars]
Remediation
## Remediation Suggestions 1. Vendor the required CLI implementation into a reviewable package or include a reproducible reference to the exact audited source revision. 2. Verify the npm artifact against an approved cryptographic integrity digest rather than relying only on a version number. 3. Commit and enforce a lockfile for all transitive dependencies where applicable. 4. Correct the `1.0.9` versus `1.2.3` documentation inconsistency and use one reviewed version throughout. 5. Review package lifecycle scripts and disable them with `--ignore-scripts` where the CLI does not require them. 6. Prefer an isolated, least-privilege environment instead of global installation. 7. Require explicit user approval before downloading or installing the package. 8. Restrict filesystem and network access to the minimum required endpoints and directories. 9. Ensure API credentials are scoped, rotated regularly, and never exposed through command-line arguments or logs. 10. Audit the package publisher, release provenance, transitive dependency tree, and npm registry configuration before deployment.

T08 · Insecure Dependencies

Warning
Location
SKILL-cn.md:48
Finding
Execution of an Unreviewed Third-Party npm Package in the Localized Skill Document## Vulnerability Details **File Location**: `SKILL-cn.md:48` **Vulnerability Type**: Third-party supply-chain dependency execution **Risk Level**: Medium **Vulnerable Code Snippet**: ```bash npx @dlazy/cli@1.2.3 <command> ``` ### Technical Analysis The localized Skill document duplicates the instruction to execute `@dlazy/cli@1.2.3` from npm. The package implementation and its transitive dependencies are not present in the audited project, so their behavior cannot be independently inspected as part of this audit. `npx` can retrieve and execute package code under the invoking user's account. Pinning the package to `1.2.3` limits version drift but does not verify the artifact's identity or protect against publisher compromise, malicious package contents, registry compromise, or vulnerable transitive dependencies. The corresponding global-install instruction can also leave the third-party component installed beyond the immediate Skill execution. As in the other Skill document, the stated provenance is inconsistent: the narrative refers to version `1.0.9`, while installation and execution use version `1.2.3`. This ambiguity makes it unclear which release was actually reviewed or intended. ### Attack Path 1. The localized Skill is selected by the user or agent. 2. Its workflow directs the agent to run the dLazy CLI. 3. `npx` downloads `@dlazy/cli@1.2.3` and unresolved dependencies from npm. 4. Package lifecycle or runtime code executes with the agent user's permissions. 5. A compromised package or dependency executes attacker-controlled logic. 6. That logic can access credentials, prompts, user-provided media, environment variables, network connectivity, and user-writable files available to the CLI process. ### Impact Assessment Exploitation could result in arbitrary code execution at the privilege level of the invoking user. Potential consequences include credential disclosure, unauthorized API usage, exposure ...[truncated 539 chars]
Remediation
## Remediation Suggestions 1. Replace direct unverified `npx` execution with a reviewed, integrity-locked artifact. 2. Publish and verify cryptographic checksums, package signatures, and reproducible build provenance. 3. Pin and audit the entire dependency tree, not only the top-level CLI version. 4. Resolve the discrepancy between versions `1.0.9` and `1.2.3`. 5. Avoid global installation unless operationally necessary. 6. Run the CLI in a sandbox or container with restricted filesystem, credential, and network access. 7. Disable unnecessary npm lifecycle scripts. 8. Obtain explicit user approval before package retrieval or installation. 9. Scope the dLazy API key to minimum required permissions and provide straightforward rotation and revocation procedures. 10. Keep the localized and primary Skill documents synchronized so that both identify the same reviewed package release and security controls.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • 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
Findings (2)

Description-Behavior Mismatch

Medium
Confidence
94% confidence
Finding
The skill is presented as a brochure-design workflow, but it instructs the agent to execute terminal commands and invoke an external CLI that can install packages, authenticate, upload local files, and send prompts to remote services. That expands the trust boundary from content design into code execution and networked side effects, creating risk of unintended command execution, package-supply-chain exposure, and local data exfiltration through referenced files.

Intent-Code Divergence

Medium
Confidence
81% confidence
Finding
The skill defines a mandatory safety/quality gate requiring layout approval before mock-up generation, but later replaces that with a generic prompt-confirmation flow for the first image. This inconsistency can cause the agent to skip the intended approval checkpoint, making it easier to trigger unintended generations, extra API calls, or uploads without the user’s specific approval of the required intermediate artifact.

VirusTotal

64/64 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.