Back to skill

Security audit

分镜视频生成 Storyboard Video

Security checks for vulnerabilities and agentic risk

Overview

This skill is review-worthy because it mixes a storyboard-to-canvas workflow with broader dLazy CLI installation, credential storage, cloud uploads, and unrelated terminal image-generation instructions.

Install only if you are comfortable running the dLazy npm CLI, storing a dLazy API key locally, and sending prompts or selected media to dLazy cloud services. Prefer one-off npx use or an isolated environment, review the @dlazy/cli package source and version, and do not grant the agent broad filesystem access beyond media you intend to upload.

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:7
Finding
Unverified Third-Party CLI Installation and Execution<![CDATA[ ## Vulnerability Details **File Location**: `SKILL.md`, lines 7 and 45–49 **Vulnerability Type**: T08: Insecure Dependencies **Risk Level**: Medium ### Vulnerable Code ```yaml metadata: {"clawdbot":{"emoji":"🤖","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","author":"dlazyai","license":"see-repo","npm":"https://www.npmjs.com/package/@dlazy/cli","configLocation":"~/.dlazy/config.json","apiEndpoints":["api.dlazy.com","files.dlazy.com"]},"openclaw":{"systemPrompt":"当你需要使用此技能时,请严格遵循此技能提供的指南进行规划和执行。你可以通过调用 dlazy CLI 的各类生成模型(如 dlazy seedream-4.5 等)来完成实际的图片渲染。注意:Windows PowerShell 中不允许使用 `&` 或 `&&` 进行命令串联或后台运行,请单独且同步地执行命令。"}} ``` ```markdown - **npm package**: `@dlazy/cli` (this Skill's install field is pinned to version `1.0.9`) - **Official website**: [dlazy.com](https://dlazy.com) If you do not want to retain a global CLI on the system, run it on demand: ```bash npx @dlazy/cli@1.2.3 <command> ``` ``` ### Technical Analysis The Skill directs the agent to install or execute the external npm package `@dlazy/cli@1.2.3`. The implementation of that package is not included in the audited project, so its runtime behavior cannot be verified from the supplied files. Although a fixed version reduces dependency drift, the instructions do not require verification through an approved lockfile, package digest, signature, or independently trusted artifact. The global installation command also creates a persistent executable outside the Skill directory. Depending on npm configuration and package contents, package lifecycle scripts and CLI code can execute with the permissions of the user running the agent. There is also a provenance inconsistency: the documentation states that the install field is pinned to `1.0.9`, while both the actual metadata and the `npx` command select `1.2.3`. This discrepancy ca ...[truncated 1438 chars]
Remediation
<![CDATA[ ## Remediation Suggestions 1. Vendor and audit the required CLI implementation, or distribute it through a controlled internal artifact repository. 2. Verify the package using a trusted lockfile and an approved integrity digest before execution. 3. Review the complete transitive dependency tree and monitor it for known vulnerabilities and ownership changes. 4. Prefer a project-local, isolated installation over `npm install -g` to limit persistence and path exposure. 5. Where compatible with the verified package, disable npm lifecycle scripts during installation and explicitly run only audited setup operations. 6. Execute the CLI in a sandbox with minimum filesystem and network access. Expose only the media files explicitly selected by the user. 7. Provide the dLazy credential through a narrowly scoped secret mechanism and prevent unrelated subprocesses from reading it. 8. Correct the provenance documentation so that every version reference consistently identifies `1.2.3`, or update all commands and metadata to the intended reviewed version. 9. Pin any transitive dependencies where possible and establish a controlled process for reviewing package upgrades. ]]>

T08 · Insecure Dependencies

Warning
Location
SKILL-cn.md:7
Finding
Unverified Third-Party CLI Installation and Execution in Duplicate Skill Definition<![CDATA[ ## Vulnerability Details **File Location**: `SKILL-cn.md`, lines 7 and 45–49 **Vulnerability Type**: T08: Insecure Dependencies **Risk Level**: Medium ### Vulnerable Code ```yaml metadata: {"clawdbot":{"emoji":"🤖","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","author":"dlazyai","license":"see-repo","npm":"https://www.npmjs.com/package/@dlazy/cli","configLocation":"~/.dlazy/config.json","apiEndpoints":["api.dlazy.com","files.dlazy.com"]},"openclaw":{"systemPrompt":"当你需要使用此技能时,请严格遵循此技能提供的指南进行规划和执行。你可以通过调用 dlazy CLI 的各类生成模型(如 dlazy seedream-4.5 等)来完成实际的图片渲染。注意:Windows PowerShell 中不允许使用 `&` 或 `&&` 进行命令串联或后台运行,请单独且同步地执行命令。"}} ``` ```markdown - **npm package**: `@dlazy/cli` (this Skill's install field is pinned to version `1.0.9`) - **Official website**: [dlazy.com](https://dlazy.com) If you do not want to retain a global CLI on the system, run it on demand: ```bash npx @dlazy/cli@1.2.3 <command> ``` ``` ### Technical Analysis This duplicate Skill definition contains the same external dependency execution path as `SKILL.md`. It instructs the environment to globally install or directly execute `@dlazy/cli@1.2.3`, but the dependency implementation is not present in the audited project. No lockfile, integrity digest, package signature, or equivalent verification mechanism is specified. The instructions therefore transfer execution trust to the npm package and its transitive dependencies. A fixed package version reduces unintended upgrades but does not protect against an already-compromised release or registry account. The documentation also incorrectly describes the install field as selecting `1.0.9`, whereas the executable commands use `1.2.3`. ### Attack Path 1. The agent loads `SKILL-cn.md` and follows its npm installation or `npx` execution instruction. 2. npm retrieves version `1.2.3` ...[truncated 1067 chars]
Remediation
<![CDATA[ ## Remediation Suggestions 1. Remove duplicated dependency metadata where possible and maintain one authoritative, reviewed installation definition. 2. Vendor and audit the CLI, or obtain it from a controlled artifact repository with immutable versioning. 3. Require integrity verification through a trusted lockfile, cryptographic digest, or package signature. 4. Avoid global installation; use a project-local isolated environment with a restricted executable path. 5. Disable package lifecycle scripts when compatible with the verified CLI and its installation requirements. 6. Sandbox execution with least-privilege filesystem, credential, and network access. 7. Make only explicitly selected media available to the CLI, rather than exposing broad filesystem access. 8. Protect the API key using a scoped secret facility and rotate it if dependency compromise is suspected. 9. Resolve the `1.0.9` versus `1.2.3` documentation inconsistency and ensure both Skill documents remain synchronized. 10. Audit and monitor all transitive dependencies before publishing future Skill versions. ]]>
Vulnerability Patterns
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • 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
Findings (8)

Description-Behavior Mismatch

Medium
Confidence
96% confidence
Finding
The metadata understates that the skill may cause terminal-based execution through the dlazy CLI, while presenting itself as simple storyboard information retrieval. Hidden or poorly disclosed execution behavior weakens user consent and can cause an agent to perform networked model operations and local configuration changes unexpectedly.

Description-Behavior Mismatch

Medium
Confidence
96% confidence
Finding
The metadata understates that the skill may cause terminal-based execution through the dlazy CLI, while presenting itself as simple storyboard information retrieval. Hidden or poorly disclosed execution behavior weakens user consent and can cause an agent to perform networked model operations and local configuration changes unexpectedly.

Context-Inappropriate Capability

High
Confidence
98% confidence
Finding
The skill's stated purpose is to convert storyboard data into a canvas video pipeline, but the later instructions redirect the agent into a stepwise terminal image-generation workflow. That introduces unrelated command execution, model invocation, and user interaction patterns that expand the skill's authority beyond its declared scope and create opportunities for unauthorized actions or data exfiltration to third-party services.

Intent-Code Divergence

Medium
Confidence
95% confidence
Finding
The documentation promises storyboard-to-canvas conversion, but later operational instructions steer the agent into a different CLI-driven workflow. This kind of internal contradiction makes the skill harder to reason about and can bypass planner safeguards that rely on the documented purpose when deciding whether to activate or trust a skill.

Description-Behavior Mismatch

Medium
Confidence
96% confidence
Finding
The manifest advertises the skill as only retrieving storyboard information, while the body actually instructs the agent to build a full JSON pipeline and invoke a canvas-modification tool. This mismatch can cause users, reviewers, or orchestration systems to grant the skill more trust or broader execution than the description justifies, reducing informed consent and making risky side effects less visible.

Intent-Code Divergence

Medium
Confidence
98% confidence
Finding
The execution section injects unrelated command-execution and image-generation instructions that conflict with the storyboard-to-canvas workflow. In practice, this can steer an agent away from the expected constrained behavior into terminal usage, iterative prompting, and remote model invocation, increasing the chance of unintended actions, data exfiltration to third-party APIs, or unsafe tool use.

Vague Triggers

Medium
Confidence
86% confidence
Finding
The trigger 'Storyboard Video Generation Pipeline' is broad and does not define boundaries, exclusions, or prerequisites. An underspecified trigger increases the likelihood that the skill activates in adjacent contexts where terminal execution, auth handling, or canvas modification are not intended, especially given the skill's broader-than-declared behavior.

Natural-Language Policy Violations

Medium
Confidence
90% confidence
Finding
The skill contains mandatory Chinese-language interaction directives without offering a user language choice or respecting the surrounding context. While not a code-execution bug, it can degrade transparency and informed consent, especially when paired with critical operational instructions, because users may not fully understand what the agent is being told to do.

Static analysis

No suspicious patterns detected.