Back to skill

Security audit

社交媒体配图设计 Social Media Design

Security checks across malware telemetry and agentic risk

Overview

The skill has a clear social-media image-generation purpose, but it asks users to install and run a third-party CLI that stores an API key and uploads selected media, with incomplete containment and inconsistent version documentation.

Review the dLazy CLI source and package version before installing, prefer the npx path or another isolated environment over the global install, use a narrowly scoped and revocable dLazy API key, and only provide media files you intend to upload to dLazy's service.

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 (2)

T08 · Insecure Dependencies

Warning
Location
SKILL.md:7
Finding
Execution of an Externally Retrieved and Unverified npm CLI## Vulnerability Details **File Location**: `SKILL.md:7-48` and mirrored content in `SKILL-cn.md:7-48` **Vulnerability Type**: `T08: Insecure Dependencies` **Risk Level**: Medium **Code Snippet**: ```yaml metadata: clawdbot: requires: bins: - npm - npx install: npm install -g @dlazy/cli@1.2.3 installAlternative: npx @dlazy/cli@1.2.3 homepage: https://github.com/dlazy-ai/cli source: https://github.com/dlazy-ai/cli npm: https://www.npmjs.com/package/@dlazy/cli configLocation: ~/.dlazy/config.json apiEndpoints: - api.dlazy.com - files.dlazy.com ``` ```bash npx @dlazy/cli@1.2.3 <command> ``` ### Technical Analysis The Skill directs the host to retrieve and execute `@dlazy/cli@1.2.3` from npm. Neither the CLI implementation nor its dependency tree is included in the audited project, so its handling of credentials, local files, lifecycle scripts, and network requests cannot be verified from the reviewed files. Pinning the package to a specific version reduces exposure to unintended future upgrades, but it does not verify package integrity or package provenance. The project provides no lockfile, cryptographic checksum, signature verification, vendored source, or sandbox policy. The global installation option also places an executable in the user's environment beyond the duration of an individual generation request. The Skill documents that the CLI receives a dLazy API key, transmits prompts to `api.dlazy.com`, and uploads selected media to `files.dlazy.com`. These operations are related to the declared functionality and are disclosed, but they increase the consequences of a supply-chain compromise because the downloaded package executes with the invoking user's permissions and can access the credentials and files exposed to it. No evidence establishes that the identified package is currently malicious. The vulnerability is the trust ...[truncated 1634 chars]
Remediation
## Remediation Suggestions 1. Vendor the exact CLI source and audited dependency graph with the Skill, or otherwise make the reviewed implementation available as part of the artifact. 2. Publish verifiable package provenance and signatures, and enforce integrity verification before execution. 3. Include a lockfile that fixes all transitive dependencies rather than pinning only the top-level package. 4. Avoid global installation. Prefer an isolated, temporary execution environment with a minimal filesystem view. 5. Disable npm lifecycle scripts where operationally possible and explicitly audit any lifecycle scripts that are required. 6. Restrict outbound network access to documented endpoints and prevent access to unrelated local files. 7. Provide the API key through a short-lived secret mechanism rather than a broadly accessible process environment. 8. Use narrowly scoped, revocable API credentials and document credential rotation procedures. 9. Require explicit user selection and confirmation before uploading each local file, and warn users not to submit secrets or unnecessary personal data. 10. Perform an independent audit of `@dlazy/cli@1.2.3` and its complete transitive dependency tree before treating the Skill as trusted.

T09 · Insecure Skill Coding Practices

Note
Location
SKILL.md:7
Finding
Contradictory CLI Version and Provenance Documentation## Vulnerability Details **File Location**: `SKILL.md:7-48` and mirrored content in `SKILL-cn.md:7-48` **Vulnerability Type**: `T09: Insecure Skill Coding Practices` **Risk Level**: Low **Code Snippet**: ```yaml install: npm install -g @dlazy/cli@1.2.3 installAlternative: npx @dlazy/cli@1.2.3 ``` ```text npm package: @dlazy/cli (the Skill's install field is pinned to version 1.0.9) ``` ```bash npx @dlazy/cli@1.2.3 <command> ``` ### Technical Analysis The metadata and executable examples use `@dlazy/cli@1.2.3`, while the provenance section states that the installation field is pinned to version `1.0.9`. These claims cannot both describe the version executed by the current Skill. Version consistency is security-relevant because reviewers may inspect source code, release notes, checksums, or dependency data for the documented version. A reviewer who relies on the `1.0.9` statement may approve a different codebase from the `1.2.3` package that the installation command actually retrieves. This inconsistency does not itself prove malicious execution, and both executable commands are pinned to the same `1.2.3` version. The risk is inaccurate provenance information and the resulting possibility of reviewing or approving the wrong release. ### Attack Path 1. A security reviewer reads the provenance section and identifies `1.0.9` as the supposedly pinned dependency. 2. The reviewer audits or approves the source, release artifacts, or dependency graph for `1.0.9`. 3. A user activates the Skill. 4. The metadata or documented `npx` command retrieves and executes `1.2.3` instead. 5. Any security-relevant differences introduced between the two releases execute without having been covered by the review. ### Impact Assessment The direct impact is loss of reliable dependency provenance and reduced effectiveness of package review. If `1.2.3` contains a vulnerability or behavior absent from the reviewed `1.0.9` re ...[truncated 351 chars]
Remediation
## Remediation Suggestions 1. Select one approved CLI version and use it consistently in metadata, installation commands, examples, and provenance text. 2. Generate all displayed version references from one canonical dependency declaration to prevent documentation drift. 3. Add automated validation that fails packaging when dependency versions differ across Skill files. 4. Record the audited package version, source commit, package digest, publication provenance, and dependency-lock digest. 5. Correct both `SKILL.md` and `SKILL-cn.md` simultaneously because they contain the same inconsistency. 6. Re-audit the version that is ultimately retained before publishing the corrected Skill.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (4)

Vague Triggers

Medium
Confidence
81% confidence
Finding
The trigger phrase 'Social Media Designer (Multi-Platform Optimization)' is broad enough that the skill could activate in loosely related design or marketing contexts without clear user intent to use this specific external-CLI workflow. Because the skill drives installation and use of a third-party SaaS CLI and uploads user-provided media to remote endpoints, overbroad activation increases the chance of unintended data exposure or unnecessary command execution.

Natural-Language Policy Violations

Medium
Confidence
87% confidence
Finding
The metadata injects a Chinese system prompt that instructs the agent to strictly follow the skill and use the dlazy CLI when needed, without any user language preference check. Forcing Chinese-language operation without opt-in can cause the agent to mis-handle user intent, consent, or safety-relevant details, especially around authentication, media upload, and command execution, making unsafe actions harder for users to detect or correct.

Vague Triggers

Medium
Confidence
92% confidence
Finding
The trigger phrase "Social Media Designer (Multi-Platform Optimization)" is broad enough to activate in many ordinary content-creation contexts without strong scoping constraints. That increases the chance the agent will invoke this skill unexpectedly, leading to unneeded installation of a third-party CLI, transmission of user prompts or local media to external services, and the skill’s embedded behavioral instructions taking precedence in unrelated tasks.

Natural-Language Policy Violations

Medium
Confidence
89% confidence
Finding
The metadata embeds a Chinese system prompt that instructs the agent to strictly follow this skill and use the dlazy CLI, without checking the user’s language preference or obtaining opt-in. This can override normal agent behavior, reduce transparency for non-Chinese users and reviewers, and make risky actions such as tool execution or external API use harder to understand or contest.

VirusTotal

VirusTotal engine telemetry is currently stale for this artifact.

View on VirusTotal

Static analysis

No suspicious patterns detected.