T05 · Unauthorized Access and Privilege Escalation
- Location
src/utils.js:70- Finding
Arbitrary Local File Read and Outbound Disclosure
- Content
View full analysis
- Remediation
View remediation
Security audit
Security checks for vulnerabilities and agentic risk
This is a coherent Volcengine image-generation skill, but its image-edit features can read arbitrary local files or internal URLs and send their contents outward.
Review before installing. Use this only in a constrained runtime with a narrowly scoped Volcengine key, fixed trusted API base URL, outbound network restrictions, spending limits, and no access to sensitive local files. Do not pass private documents, secrets, internal URLs, or regulated images as image inputs.
src/utils.js:70Arbitrary Local File Read and Outbound Disclosure
src/image-gen.js:103Unrestricted API Base URL Can Exfiltrate Credentials and Request Data
src/utils.js:41Server-Side Request Forgery Through Caller-Controlled Image URLs
src/image-gen.js:235Unbounded Inputs Permit Resource Exhaustion and Excessive API Charges
YARA rule matched a known malware signature (reverse shell, backdoor, ransomware, C2 framework, or info stealer).
�控制** - 批量生成时控制并发数,避免限流
cd /home/node/openclaw-skills/volc-image-gen
npm install
Linux/macOS:
# 添加到 ~/.bashrc 或 ~/.zshrc
echo 'export VOLC_API_KEY="your_api_key_here"' >> ~/.bashrc
echo 'export VOLC_API_BASE="https://ark.cn-beijing.volces.com/api/v3"' >> ~/.bashrc
source ~/.bashrc
Windows PowerShell:
[System.Environment]::SetEnvironmentVariable("VOLC_API_KEY", "your_api_key_here", "User")
[System.Environment]::SetEnvironmentVariable("VOLC_API_BASE", "https://ark.cn-beijing.volces.com/api/v3", "User")
The README describes image generation and editing features but does not clearly disclose that user prompts and input images are transmitted to a third-party Volcengine API. In a skill that may process sensitive prompts or private images, this omission can mislead users about data handling and create privacy, compliance, and consent risks.
The skill explicitly supports sending user prompts and input images to the Volcano Engine third-party API, but the documentation does not clearly warn that those inputs leave the local environment and may contain sensitive personal, proprietary, or regulated data. This can lead users to unknowingly transmit confidential images or prompts to an external service, creating privacy, compliance, and data-governance risk.
The alias set for the image-generation command includes very generic triggers such as "img", "image", and common Chinese phrases for generating pictures. In agent environments that route on simple alias matching, broad triggers can cause accidental invocation from ordinary user text, leading to unintended API calls, cost incurrence, or side effects like image generation and downstream file handling.
The manifest advertises automatic local image downloads but provides no warning, path restrictions, or indication of user consent around filesystem effects. In a skill that processes remote-generated content, automatic writes can create unwanted local persistence, fill disk space, or save untrusted files in sensitive locations if implementation controls are weak.
This file's natural-language content, including comments, errors, and console output, is entirely in Chinese, and there is no indication that users can choose another language or that the skill is intentionally region- or locale-specific. The policy requires flagging language or locale constraints when they are imposed without user opt-in or justification.
This JavaScript file uses Chinese-only natural-language comments and user-visible status/error strings throughout, beginning with the module description and continuing in returned errors and logs. Under the policy, forcing a specific language without an explicit user choice is a locale/language policy violation unless the constraint is clearly documented and justified.
The function sends the composed prompt and generation parameters to the Volcengine Ark API using axios.post. While the module comment says it calls the API, there is no explicit warning that user-supplied prompt content will be transmitted to a third-party service, which is relevant for privacy-sensitive prompts.
The code automatically downloads generated images to a local path via downloadImage(img.url, filename), which is a file-write operation. Although there are progress logs, they do not clearly disclose beforehand that files will be saved locally, and the inline comments/docstrings do not warn users about this storage side effect.
The user-facing descriptions, errors, help text, and primary command aliases are written in Chinese throughout the file, with no indication that other languages are supported or that the language is configurable. This is a natural-language policy concern because the skill effectively imposes a specific language/locale on users without opt-in.
This file's natural-language content is entirely in Chinese, including module documentation and user-facing error/log strings. Under the stated policy, forcing a specific language without user opt-in can be a locale-policy violation when no justification or alternative language path is provided.
The utility exposes arbitrary local file reading by accepting either a URL or any existing local path and then base64-encoding the file contents without restriction. If untrusted input can reach loadImage(), an attacker could read sensitive local files such as credentials, keys, or application data, making this a real file disclosure issue rather than a harmless helper.
The documentation advertises caching and later examples mention local paths, but it does not clearly warn that generated images and cached results may be written to local disk. This can expose sensitive generated content to other local users, backups, or disk forensics if operators assume processing is memory-only.
The description is entirely in Chinese and does not indicate that users can choose another language or that the skill is intentionally limited to a Chinese-speaking or China-region audience. This creates a natural-language locale-policy concern because it implicitly fixes the skill's presentation to one language without opt-in.
Using a caret range for axios allows future installs to resolve to different patch/minor releases, reducing build reproducibility and making it harder to verify whether a safe version is installed. In a network-facing library like axios, this also increases supply-chain risk because vulnerable or behavior-changing releases may be pulled in unexpectedly.
"author": "OpenClaw Skill Master",
"license": "MIT",
"dependencies": {
"axios": "^1.6.0",
"p-limit": "^3.1.0",
"node-cache": "^5.1.2"
},
The manifest uses axios without pinning an exact version, and axios has multiple known advisories in some releases. Because the resolved installed version is not fixed here, it is impossible to verify from this file alone whether deployments may include a vulnerable axios release, which is significant in a skill likely to make outbound API requests.
Using a caret range for p-limit makes dependency resolution non-deterministic across environments and weakens supply-chain assurance. Although p-limit is lower risk than an HTTP client, unpinned dependencies can still introduce malicious or vulnerable code through later published versions.
"license": "MIT",
"dependencies": {
"axios": "^1.6.0",
"p-limit": "^3.1.0",
"node-cache": "^5.1.2"
},
"engines": {
Using a caret range for node-cache means installations may silently pick up different releases over time, which undermines reproducibility and dependency integrity. While the direct impact is limited in this manifest alone, it is still a supply-chain hygiene weakness.
"dependencies": {
"axios": "^1.6.0",
"p-limit": "^3.1.0",
"node-cache": "^5.1.2"
},
"engines": {
"node": ">=18.0.0"
The help command exposes the alias "帮助"/"usage", which is a common everyday phrase and may be triggered unintentionally if the platform matches aliases loosely. While lower risk than content-generating commands, unintended activation can still leak capability details, confuse routing, or create noisy behavior that makes abuse and phishing-style prompt steering easier.
The file’s human-facing text, including the header comment and all console output, is written exclusively in Chinese. This creates a natural-language locale constraint without any opt-in, fallback, or indication that the skill is intentionally region-specific.