Back to skill

Security audit

Tiexue Vision

Security checks for vulnerabilities and agentic risk

Overview

The skill is not deceptive, but it should be reviewed because it automatically writes recognition results back and may send OCR text to Google for translation without per-use consent.

Install only if you are comfortable with images in Feishu chats/docs being processed automatically, results being written back or saved locally, and English/ASCII OCR text being sent to Google Translate. Prefer a version that requires explicit invocation, offers preview before edits, supports local-only mode, and updates/pins vulnerable dependencies.

Vulnerability Patterns
  • 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
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
Findings (1)

T09 · Insecure Skill Coding Practices

Warning
Location
index.js:36
Finding

OCR Content Is Disclosed to an External Translation Service Through URL Query Parameters

Content
View full analysis

Vulnerability Details

File Location: index.js, lines 36–49
Vulnerability Type: Automatic external disclosure of potentially sensitive OCR content
Risk Level: Medium

Vulnerable Code

js
async function translateIfEnglish(text) {
  const isEnglish = /^[\x00-\x7F]*$/.test(text);
  if (!isEnglish) return text;
  const resp = await axios.get('https://translate.googleapis.com/translate_a/single', {
    params: {
      client: 'gtx',
      sl: 'en',
      tl: 'zh-CN',
      dt: 't',
      q: text
    }
  });
  return resp.data[0].map(item => item[0]).join('');
}

Technical Analysis

OCR output is automatically transmitted to translate.googleapis.com whenever the entire value consists of ASCII characters. The regular expression is not a reliable English-language detector: it also accepts passwords, API tokens, identifiers, financial data, source code, numbers, and empty strings.

Axios serializes q: text into the GET request's URL query string. Although HTTPS protects the request in transit, URLs can be recorded by the destination service and by client, proxy, gateway, observability, or security logging systems. This creates broader exposure than placing sensitive content in a controlled request body.

The documentation discloses that translation is an exception to local processing, so the behavior is not hidden exfiltration. Nevertheless, the implementation does not obtain per-request consent, classify or redact sensitive OCR data, or provide a local-only enforcement control.

Attack Path

  1. A user submits an image containing sensitive ASCII text, such as a password, API token, customer record, or confidential internal document.
  2. ocrImage() extracts that content as text.
  3. translateIfEnglish() treats the content as English solely because it contains only ASCII characters.
  4. Axios inserts the full OCR output into the q URL query parameter.
  5. The request is sent to Google's translation endpoint.
  6. The sensitive conten ...[truncated 765 chars]
Remediation
View remediation

Remediation Suggestions

  1. Require explicit user or administrator opt-in before sending OCR content to any external translation service.
  2. Provide and enforce a local-only mode in which outbound translation requests are prohibited.
  3. Replace the ASCII test with a proper language-detection mechanism, and do not translate empty, numeric-only, token-like, or identifier-like content.
  4. Scan and redact likely credentials, tokens, personal data, and other sensitive values before external transmission.
  5. Use a documented, authenticated translation API whose privacy, retention, regional processing, and data-handling terms are appropriate for the deployment.
  6. Prefer a POST request body instead of a GET query parameter where supported, while recognizing that this alone does not eliminate disclosure to the service.
  7. Apply request-size limits, timeouts, explicit destination allowlisting, and redirect restrictions.
  8. Clearly disclose the destination and transmitted fields before processing, and obtain consent appropriate to the data's sensitivity.
  9. Avoid logging raw OCR content or complete translation requests in application and infrastructure logs.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • YARA SignaturesMalware Match, Webshell Match, Cryptominer Match
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (22)

Known Vulnerable Dependency: tar==6.2.1 — 12 advisory(ies): CVE-2026-59873 (node-tar: Decompression/parse DoS via unlimited input); CVE-2026-24842 (node-tar Vulnerable to Arbitrary File Creation/Overwrite via Hardlink Path Trave); CVE-2026-26960 (Arbitrary File Read/Write via Hardlink Target Escape Through Symlink Chain in no) +9 more

Critical
Category
Supply Chain
Confidence
97% confidence
Finding

tar 6.2.1 is associated with multiple critical archive-handling issues, including path traversal, arbitrary file overwrite/creation, and denial of service. This is especially concerning here because the lockfile contains several packages with install scripts and native/runtime components that commonly download or unpack binary artifacts, so vulnerable tar processing can directly affect installation or runtime artifact handling.

Content

No source excerpt is available for this finding.

Known Vulnerable Dependency: adm-zip==0.5.17 — 2 advisory(ies): CVE-2026-76845 (adm-zip extraction follows destination symlinks, allowing arbitrary file overwri); CVE-2026-39244 (adm-zip: Crafted ZIP file triggers 4GB memory allocation)

High
Category
Supply Chain
Confidence
94% confidence
Finding

adm-zip 0.5.17 is flagged with high-severity issues involving unsafe ZIP extraction behavior and memory exhaustion on crafted archives. In this dependency set, the risk is credible because multiple installed packages with native-model/runtime behavior may download or unpack archives during install or runtime, making archive-processing bugs materially relevant rather than theoretical.

Content

No source excerpt is available for this finding.

Known Vulnerable Dependency: axios==1.14.0 — 16 advisory(ies): CVE-2026-44494 (axios Vulnerable to Full Man-in-the-Middle via Prototype Pollution Gadget in `co); CVE-2026-44495 (axios Vulnerable to Credential Theft and Response Hijacking via Prototype Pollut); CVE-2025-62718 (Axios has a NO_PROXY Hostname Normalization Bypass that Leads to SSRF) +13 more

High
Category
Supply Chain
Confidence
90% confidence
Finding

axios 1.14.0 is reported with multiple advisories including SSRF/proxy-bypass and prototype-pollution-related request manipulation. This matters more in this skill because the top-level dependencies include both axios and node-fetch, indicating outbound network access is part of the design, so HTTP client flaws can become exploitable when attacker-controlled URLs, headers, or proxy settings are involved.

Content

No source excerpt is available for this finding.

Known Vulnerable Dependency: brace-expansion==1.1.13 — 3 advisory(ies): CVE-2026-13149 (brace-expansion: DoS via exponential-time expansion of consecutive non-expanding); CVE-2026-14257 (brace-expansion: DoS via unbounded expansion length causing an out-of-memory pro); CVE-2026-69152 (brace-expansion: DoS via unbounded intermediate arrays, bypassing the CVE-2026-1)

High
Category
Supply Chain
Confidence
82% confidence
Finding

brace-expansion 1.1.13 has reported denial-of-service issues from pathological expansion inputs that can trigger extreme CPU or memory usage. As a transitive dependency used for pattern/glob handling, the vulnerability is real, though exploitability depends on whether untrusted input can reach code paths that parse brace patterns.

Content

No source excerpt is available for this finding.

Known Vulnerable Dependency: form-data==4.0.5 — 1 advisory(ies): CVE-2026-12143 (form-data: CRLF injection in form-data via unescaped multipart field names and f)

High
Category
Supply Chain
Confidence
85% confidence
Finding

form-data 4.0.5 is reported vulnerable to CRLF injection through unescaped multipart field names or filenames, which can corrupt HTTP requests or inject unintended headers/body content. Given this skill includes HTTP client libraries and likely handles file/model/image uploads, malformed attacker-controlled metadata could plausibly reach multipart construction paths.

Content

No source excerpt is available for this finding.

YARA rule 'agent_skill_remote_bootstrap_execution': Remote script or code download followed by execution/bootstrap installation [agent_skills]

High
Category
YARA Match
Confidence
85% confidence
Finding

YARA rule matched a known malware signature (reverse shell, backdoor, ransomware, C2 framework, or info stealer).

Content

Scanner excerpt · package.json (reported line 8)May include surrounding context.

json
{
  "name": "openclaw-vision-skill",
  "version": "1.0.0",
  "description": "图片文字+物体+场景全能识别并写回",
  "main": "index.js",
  "type": "module",
  "scripts": {
    "postinstall": "node -e \"console.log('依赖已安装,准备使用')\""
  },
  "dependencies": {
    "@tensorflow/tfjs-node": "^4.12.0",
    "onnxruntime-node": "^1.15.0",
    "tesseract.js": "^5.0.5",
    "axios": "^1.7.2",
    "node-fetch": "^3.3.2"
  }
}

Known Vulnerable Dependency: axios==1.14.0 — 16 advisory(ies): CVE-2026-44494 (axios Vulnerable to Full Man-in-the-Middle via Prototype Pollution Gadget in `co); CVE-2026-44495 (axios Vulnerable to Credential Theft and Response Hijacking via Prototype Pollut); CVE-2025-62718 (Axios has a NO_PROXY Hostname Normalization Bypass that Leads to SSRF) +13 more

High
Category
Supply Chain
Confidence
97% confidence
Finding

The manifest includes axios with a version range that static analysis associates with known advisories, including SSRF/proxy-handling classes of issues. In a skill that appears designed to perform recognition and likely communicate over the network, a vulnerable HTTP client materially raises the chance of internal network access, credential exposure, or tampered responses.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The skill description mentions write-back behavior, but the warning is not prominent relative to the automation described, and users may not fully appreciate that outputs will be inserted into original chats/documents or created as local files. This increases the risk of accidental disclosure, document corruption, or persistence of sensitive OCR results in places users did not intend.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

Automatically translating detected English text into Chinese and writing it back without user choice can alter the meaning of extracted content and overwrite or replace the original language unexpectedly. This is especially risky for legal, technical, or security-relevant text where mistranslation could mislead users or damage the integrity of the source content.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The skill is documented to auto-trigger on any image sent in chats or documents, which is overly broad and can cause unintended processing of user content. In a collaborative environment, this can lead to surprise data handling, unintended OCR of sensitive images, and automatic write-back into shared contexts without explicit per-use consent.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The skill sends OCR-extracted text to translate.googleapis.com whenever the text is ASCII, which can disclose potentially sensitive contents from user images to a third-party service without explicit consent or notice. Because OCR may capture personal, financial, or confidential information, this creates a real privacy and data-handling risk even though it is not an exploit in the traditional code-execution sense.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The code hardcodes translation target language to zh-CN and automatically translates English text without user choice. This is primarily a privacy and integrity issue: users may not expect automatic transformation of extracted content, and forced translation can alter meaning or mishandle sensitive text in workflows that require fidelity.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
84% confidence
Finding

This manifest file contains only a broad capability description ('image text + object + scene recognition and write back') without any explicit trigger phrases, invocation constraints, or negative examples. In a manifest context, the absence of specific activation guidance can make it unclear when the skill should be invoked versus not invoked.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The natural-language description is presented only in Chinese, indicating a language-specific skill description without any visible user choice or opt-in. This can violate language/locale policy when the skill does not clearly offer multilingual support or document that it is intended only for a Chinese-speaking context.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
86% confidence
Finding

The config hard-codes OCR processing to English and Simplified Chinese via the tesseractLang setting. This is a natural-language/locale policy concern because it constrains language handling without any indication of user selection, opt-in, or documented justification in the file.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
80% confidence
Finding

When the source is not Feishu, the code unconditionally writes results to a new file beside the input image. Although this is part of the flow, there is no visible prompt, log, comment, or docstring disclosing that a file will be created.

Content

No source excerpt is available for this finding.

Known Vulnerable Dependency: follow-redirects==1.15.11 — 1 advisory(ies): CVE-2026-40895 (follow-redirects leaks Custom Authentication Headers to Cross-Domain Redirect Ta)

Low
Category
Supply Chain
Confidence
79% confidence
Finding

follow-redirects 1.15.11 is flagged for leaking custom authentication headers across cross-domain redirects. This is a real issue in networked applications, but impact is lower unless the skill sends sensitive headers and follows attacker-influenced redirects to untrusted domains.

Content

No source excerpt is available for this finding.

Unpinned Dependencies

Low
Category
Supply Chain
Confidence
90% confidence
Finding

Using caret ranges for @tensorflow/tfjs-node allows automatic uptake of future minor/patch releases, which can unexpectedly introduce vulnerable or incompatible code into the supply chain. In an agent skill context, dependency drift is risky because native/ML packages often have large attack surfaces and postinstall/native binary behavior.

Content

Scanner excerpt · package.json (reported line 11)May include surrounding context.

json
"postinstall": "node -e \"console.log('依赖已安装,准备使用')\""
  },
  "dependencies": {
    "@tensorflow/tfjs-node": "^4.12.0",
    "onnxruntime-node": "^1.15.0",
    "tesseract.js": "^5.0.5",
    "axios": "^1.7.2",

Unpinned Dependencies

Low
Category
Supply Chain
Confidence
90% confidence
Finding

Using a version range for onnxruntime-node weakens build reproducibility and increases exposure to compromised or vulnerable transitive updates. Because this package includes native runtime components, unexpected version changes can carry elevated supply-chain risk.

Content

Scanner excerpt · package.json (reported line 12)May include surrounding context.

json
},
  "dependencies": {
    "@tensorflow/tfjs-node": "^4.12.0",
    "onnxruntime-node": "^1.15.0",
    "tesseract.js": "^5.0.5",
    "axios": "^1.7.2",
    "node-fetch": "^3.3.2"

Unpinned Dependencies

Low
Category
Supply Chain
Confidence
88% confidence
Finding

The unpinned tesseract.js dependency permits silent dependency changes over time, making the skill harder to audit and increasing supply-chain exposure. While not immediately exploitable on its own, it weakens integrity controls for a package that may process untrusted content.

Content

Scanner excerpt · package.json (reported line 13)May include surrounding context.

json
"dependencies": {
    "@tensorflow/tfjs-node": "^4.12.0",
    "onnxruntime-node": "^1.15.0",
    "tesseract.js": "^5.0.5",
    "axios": "^1.7.2",
    "node-fetch": "^3.3.2"
  }

Unpinned Dependencies

Low
Category
Supply Chain
Confidence
98% confidence
Finding

The axios dependency is both unpinned and separately flagged as vulnerable, so allowing floating updates compounds existing risk. In a vision skill that likely fetches remote content or APIs, HTTP client weaknesses can enable SSRF, proxy bypass, credential leakage, or response manipulation.

Content

Scanner excerpt · package.json (reported line 14)May include surrounding context.

json
"@tensorflow/tfjs-node": "^4.12.0",
    "onnxruntime-node": "^1.15.0",
    "tesseract.js": "^5.0.5",
    "axios": "^1.7.2",
    "node-fetch": "^3.3.2"
  }
}

Unpinned Dependencies

Low
Category
Supply Chain
Confidence
87% confidence
Finding

node-fetch is specified with a caret range, which creates supply-chain uncertainty and undermines reproducibility for a network-capable package. This is especially relevant in agent skills that may retrieve remote content based on external inputs.

Content

Scanner excerpt · package.json (reported line 15)May include surrounding context.

json
"onnxruntime-node": "^1.15.0",
    "tesseract.js": "^5.0.5",
    "axios": "^1.7.2",
    "node-fetch": "^3.3.2"
  }
}