Back to skill

Security audit

My skill

Security checks for vulnerabilities and agentic risk

Overview

The skill is mostly a simple utility plugin, but it needs Review because its MCP tool can make unrestricted network requests from the user's OpenClaw environment.

Install only if you are comfortable with an MCP client tool that can contact arbitrary URLs from inside your OpenClaw environment. Prefer trusted MCP endpoints, avoid internal or sensitive network addresses, and pin the package to the reviewed version before installation.

Vulnerability Patterns
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
  • 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
Findings (3)

T05 · Unauthorized Access and Privilege Escalation

Error
Location
index.ts:397
Finding

Agent-Callable Server-Side Request Forgery Through Arbitrary MCP URLs

Content
View full analysis
= {}): Promise { const urlObj = new URL(url); const isHttps = urlObj.protocol === "https:"; const httpModule = isHttps ? (await import("https")) : http; return new Promise((resolve, reject) => { const body = JSON.stringify({ jsonrpc: "2.0", id: 1, method, params }); const options = { hostname: urlObj.hostname, port: urlObj.port || (isHttps ? 443 : 80), path: urlObj.pathname, method: "POST", headers: { "Content-Type": "application/json", "Content-Length": Buffer.byteLength(body) } }; const req = httpModule.request(options, (res) => { let data = ""; res.on("data", chunk => { data += chunk; }); res.on("end", () => { try { const response = JSON.parse(data); resolve(response.result || response); } catch (e) { reject(e); } }); }); req.on("error", reject); req.write(body); req.end(); }); } ``` ```ts const url = params.url || "http://localhost:3000"; try { const result = await mcpRequest(url, "initialize", { protocolVersion: "2024-11-05", capabilities: {}, clientInfo: { name: "openclaw-mcp", version: "1.0.0" } }) as { serverInfo: { name: string; version: string } }; ``` ### Technical Analysis The `mcp_echo` tool accepts a caller-controlled URL and passes it directly to a server-side HTTP client. The implementation performs no destination allowlisting, hostname validation, DNS resolution validation, private-address filtering, or protocol restriction beyond treating `https:` specially. Any other parsed protocol is effectively sent through the HTTP module. Because the request originates from the OpenClaw host rather than the exte ...[truncated 1636 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
index.ts:412
Finding

Unbounded MCP Response Buffering and Missing Network Timeouts

Content
View full analysis
{ let data = ""; res.on("data", chunk => { data += chunk; }); res.on("end", () => { try { const response = JSON.parse(data); resolve(response.result || response); } catch (e) { reject(e); } }); }); req.on("error", reject); req.write(body); req.end(); ``` ### Technical Analysis The MCP client appends every response chunk to an in-memory string and parses the response only after the server ends the stream. There is no maximum response size, so a server can force the process to retain an arbitrarily large body. The request also has no connection, inactivity, or total execution timeout. A server can accept the connection and then respond indefinitely, send data extremely slowly, or never complete the response. Each request can consequently retain sockets, promises, buffers, and execution capacity for an uncontrolled period. The arbitrary destination accepted by `mcp_echo` makes exploitation practical: an attacker who can influence a tool call can direct the client to an attacker-controlled server. ### Attack Path 1. The attacker operates an HTTP server that accepts the plugin's POST request. 2. The attacker causes the Agent to invoke `mcp_echo` with that server's URL. 3. The server either streams an unlimited response body, sends data slowly without ending the response, or accepts the connection and remains idle. 4. The plugin continues accumulating response data or waiting because it has no byte limit or timeout. 5. One large response may cause excessive memory consumption; repeated requests can exhaust memory, sockets, or available concurrent execution capacity. 6. The OpenClaw process may become slow, unavailable, or terminate due to resource exhaustion. ### Impact Assessment Exploitation can d ...[truncated 310 chars]
Remediation
View remediation

T08 · Insecure Dependencies

Note
Location
SKILL.md:44
Finding

Documentation Installs a Mutable Unpinned Plugin Release

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (10)

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The skill advertises dynamic skill loading and MCP tool invocation without any warning that these features may import or execute externally provided capabilities. In an agent ecosystem, this increases the risk of prompt-injection, execution of untrusted tool behaviors, or unintended access to external systems because users are not alerted to trust boundaries.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
86% confidence
Finding

This README lists config_set, config_export, and config_reset as available actions, including a reset operation, but provides no warning that these tools can alter or clear configuration state. For markdown files, safety-relevant behaviors affecting user data or system integrity should be disclosed so users understand the impact before use.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The MCP parameter schema and top-level comments advertise both stdio and http server types and list tooling for MCP server interaction. However, the implementation explicitly says only HTTP is implemented, ignores type, command, and args, and mcp_echo does not store a server in mcpServers, making later server-oriented tools unusable as documented.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The tool description says it will dynamically load a specified SKILL.md file, which implies reading and loading actual skill content. The implementation only looks up params.skillName in a static skills array and returns formatted metadata, without loading any file or skill document.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

The tool description states that it sets a configuration item's value. In reality, it only echoes the requested change and tells the caller to use another gateway command for actual modification, which directly contradicts the documented intent.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The description says the tool exports the current configuration, which implies returning real configuration data. The implementation instead returns a placeholder message instructing users to use external commands, so the documented behavior is not what the code does.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The tool claims it resets configuration values to defaults, which is a state-changing operation. The code only reports what would be reset and directs the user elsewhere for real behavior, contradicting the stated intent.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

mcpRequest sends arbitrary JSON-RPC requests to user-supplied URLs with no allowlist, no destination restrictions, and no user-facing disclosure that data will be transmitted over the network. In an agent/plugin context, this creates SSRF-style risk and can expose prompts, tool arguments, or internal metadata to attacker-controlled endpoints, especially if localhost or internal network targets are reachable.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

mcp_call_tool forwards arbitrary caller-provided arguments to a remote MCP server and returns the response without meaningful trust boundaries or disclosure. In the skill context, this is more dangerous because the tool can become a generic exfiltration and remote-action proxy: sensitive workspace data or agent-derived content could be packaged into arguments and sent to an untrusted external service.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
82% confidence
Finding

Natural-language strings throughout the file, including introductory comments and many tool/parameter descriptions, are presented only in Chinese with no option for user language choice. This can violate language/locale policy when the skill does not document that it is intentionally region-specific or provide an opt-in mechanism.

Content

No source excerpt is available for this finding.