Back to skill

Security audit

Byted Security Llmscanner

Security checks for vulnerabilities and agentic risk

Overview

This skill has a coherent security-testing purpose, but its implementation can expose credentials and includes unsafe command and file handling that warrants review before installation.

Install only in a controlled environment after reviewing and fixing the TLS bypasses, task ID validation, path containment, and secret handling. Use test or least-privilege platform accounts, avoid pasting real API keys into chat or command lines, rotate any keys used with this version, and verify npm dependencies before running mutating asset or task commands.

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 (5)

T09 · Insecure Skill Coding Practices

Error
Location
scripts/src/run_analysis.ts:24
Finding

Shell Command Injection Through Unvalidated Task ID

Content
View full analysis
{ console.log('📂 检测到本地已有任务数据,开始分析...'); try { const { stdout } = await execAsync(`npx ts-node src/analyze_task_data.ts ${taskId}`); console.log(stdout); return true; } catch (error: any) { console.error(`❌ 分析失败:${error.stderr || error.message}`); return false; } } ``` The value reaches this function directly from command-line input: ```ts const args = process.argv.slice(2); if (args.length !== 1) { // ... process.exit(1); } const taskId = args[0]; ``` ### Technical Analysis `taskId` is inserted directly into a command string passed to Node.js `child_process.exec`. The `exec` API invokes a shell, so shell metacharacters, command substitutions, separators, and redirections contained in `taskId` are interpreted by that shell. There is no strict Task ID validation, shell escaping, or use of an argument-array process API. Although normal platform Task IDs appear to be hexadecimal identifiers, the implementation does not enforce that format. Exploitation requires the execution flow to reach `analyzeLocalData`. This happens when a matching local data file exists or after data for a successfully completed task is fetched and saved. ### Attack Path 1. An attacker causes the Skill to invoke `run_analysis.ts` with a malicious Task ID containing shell syntax. 2. The local-data condition is satisfied, for example through a pre-positioned matching data file, or another execution path reaches `analyzeLocalData`. 3. `analyzeLocalData` concatenates the Task ID into: ```sh npx ts-node src/analyze_task_data.ts ``` 4. `exec` passes the string to a shell. 5. The shell interprets the injected syntax and executes the attacker-supplied command. ### ...[truncated 419 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
scripts/src/common.ts:174
Finding

Path Traversal Through Task-ID-Derived File Paths

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
scripts/src/common.ts:96
Finding

TLS Certificate Verification Disabled for Sensitive API Requests

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
scripts/src/common.ts:81
Finding

Platform Credentials and Bearer Tokens Stored in Plaintext

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
scripts/src/create_asset_agent.ts:133
Finding

Agent API Keys Exposed Through Process Arguments and Console Logs

Content
View full analysis
0) { console.log(` ChatVars :${JSON.stringify(chatVars)}`); } ``` The documented invocation explicitly encourages this pattern: ```sh npx ts-node src/create_asset_agent.ts my-agent \ b0226c4550aa4791a8c19a118a5f8ef5 \ --api_key=sk-xxx ``` ### Technical Analysis Secrets passed through command-line arguments can be retained in shell history and may be visible in process inspection facilities while the command is running. The program then serializes and prints the complete `ChatVars` object, causing a second disclosure through terminal capture, OpenClaw output, CI logs, or centralized logging. The same behavior exists in both agent creation and agent update scripts. Transmission of agent credentials to the configured assessment platform is part of the declared asset-management function, but exposing those values through argv and logs is unnecessary. ### Attack Path 1. A user follows the documented invocation and supplies an agent API key through `--api_key=`. 2. The complete command is retained in shell history or exposed through process inspection. 3. The script parses the key into `chatVars`. 4. The script logs `JSON.stringify(chatVars)` without redaction. 5. A local observer or anyone with access to re ...[truncated 417 chars]
Remediation
View remediation
[ key, /key|token|secret|password|credential/i.test(key) ? '[REDACTED]' : value ]) ); ``` - Prefer logging only the names of configured variables, not their values. - Apply the same remediation to both creation and update scripts. - Update documentation to remove examples that place real secrets in command lines. - Advise affected users to clear exposed shell and CI logs and rotate keys that may already have been recorded. ]]>
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Tool MisuseTool Parameter Abuse, Chaining Abuse, Unsafe Defaults
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
Findings (119)

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
80% confidence
Finding

The troubleshooting guidance recommends using curl with -k, which disables TLS certificate verification. Normalizing insecure transport checks can train users to ignore certificate failures and may expose credentials or session data to man-in-the-middle interception if reused beyond basic diagnostics.

Content

Scanner excerpt · README.md (reported line 495)May include surrounding context.

  1. 确认 host 地址是否可访问
  2. 尝试使用 curl 测试 API 连接:
    bash
    curl -k https://你的API地址:端口/api/top/wuji/cn-north-1/2025-01-01/Login
    

问题 2:助手提示认证失败或权限错误

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
85% confidence
Finding

Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).

Content

Scanner excerpt · README.md (reported line 505)May include surrounding context.

解决方法:

  • 删除 token 缓存文件,系统会自动重新获取:
    bash
    rm ~/.openclaw/workspace/skills/byted-security-llmscanner/scripts/data/token_cache.json
    

问题 3:助手提示找不到测评集或资产

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 30)May include surrounding context.

md
- **`update_asset_model.ts`** 的第一个参数是 `Name`(资产名称),不是 `AssetID`。脚本会自动根据名称和ModelID查询资产ID。

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 31)May include surrounding context.

md
- **`update_asset_agent.ts`** 的参数是 `<AgentID> <Name> <PlatformID>`,需要同时提供资产ID和名称。

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 296)May include surrounding context.

md
- `ScenarioID`:安全测评剧本 ID(通过 `list_rt_scenarios.ts` 查询)

Known Vulnerable Dependency: axios==1.13.6 — 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
93% confidence
Finding

The lockfile pins axios 1.13.6, and the supplied advisory set includes multiple high-severity issues such as SSRF/proxy-bypass and prototype-pollution-related request/response manipulation. In a security assessment tool that likely makes outbound HTTP requests to model or evaluation services, a vulnerable HTTP client is especially risky because it can enable request redirection, credential leakage, or abuse of trusted network paths.

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
88% confidence
Finding

form-data 4.0.5 is reported as vulnerable to CRLF injection through unescaped multipart field names/filenames. If this tool uploads files or constructs multipart requests using untrusted input, an attacker may be able to manipulate part boundaries or inject unintended headers/content into downstream HTTP requests.

Content

No source excerpt is available for this finding.

Known Vulnerable Dependency: axios==1.13.6 — 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 1.13.6, which the finding states is affected by multiple known advisories including SSRF-related and prototype-pollution-assisted attack scenarios. In a security assessment tool that likely makes outbound HTTP requests to model, agent, or evaluation services, a vulnerable HTTP client can materially increase risk by enabling request routing bypass, credential leakage, or response manipulation depending on usage.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The login request explicitly disables TLS certificate verification via rejectUnauthorized: false, which allows man-in-the-middle interception or spoofing of the authentication endpoint. In this skill, that is especially dangerous because the same request carries username/password credentials and yields an authentication token, so an attacker on the network could steal credentials or issue a fake token response.

Content

No source excerpt is available for this finding.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The script explicitly sets rejectUnauthorized: false on HTTPS requests, disabling TLS certificate validation. That allows man-in-the-middle interception or spoofing of the remote API, which is especially dangerous here because the script sends bearer tokens and performs authenticated task-management actions against a security assessment platform.

Content

No source excerpt is available for this finding.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The HTTPS client sets rejectUnauthorized: false, disabling certificate validation and allowing man-in-the-middle interception or spoofing of the API endpoint. Because this script handles bearer tokens and task status for a security-evaluation platform, an attacker on the network could steal credentials, tamper with task metadata, or redirect requests to a malicious server.

Content

No source excerpt is available for this finding.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

This paginated data retrieval path also disables TLS certificate verification, exposing the bulk task results to interception and manipulation. Since the code downloads and later analyzes evaluation data, a network attacker could alter result content, poison downstream analysis, or exfiltrate sensitive assessment artifacts.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The code creates an HTTPS agent with rejectUnauthorized: false, which disables TLS certificate validation for the asset update request. This allows a man-in-the-middle attacker to intercept or tamper with the authenticated API call, potentially stealing the bearer token and sensitive ChatVars such as API keys, or modifying the asset update payload.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The README instructs users to provide API keys and login credentials directly in chat examples, but gives no warning that conversational inputs may be logged, retained, or exposed to the agent/runtime. This creates a real secret-handling risk because users may copy production credentials into a channel not designed for secure secret entry.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · README.md (reported line 144)May include surrounding context.

md
**高级用法示例:**

> 你:帮我创建一个大模型测评资产,名称是 gpt-4-vision,模型 ID 是 gpt-4-vision-preview,API 地址是 https://api.openai.com/v1,API 密钥是 sk-xxxxxxxxxxxx,支持文生文和图文生文

---

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · README.md (reported line 232)May include surrounding context.

md
**高级用法示例(带 JSON 配置):**

> 你:帮我创建一个 Dify 智能体测评资产,名称是代码助手,平台 ID 是 b0226c4550aa4791a8c19a118a5f8ef5,配置是 {"url":"https://api.dify.ai/v1","api_key":"app-xxxxxxxxxxxx"}

---

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The documentation states that reports and cache files are written to disk, but does not warn that they may contain sensitive evaluation data, attack payloads, model outputs, or tokens. On shared systems or poorly permissioned environments, these artifacts could leak proprietary prompts, security findings, or credentials.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The trigger conditions are broad keyword matches such as '创建资产', '更新资产', and '发起测评', which can overlap with ordinary discussion and cause the skill to activate in unintended contexts. Because this skill drives operational actions around assets and evaluation tasks, accidental invocation could lead to unintended task creation, asset modification, or exposure of follow-on prompts requesting sensitive parameters.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.