Back to skill

Security audit

humancanhelp

Security checks for vulnerabilities and agentic risk

Overview

This skill is a disclosed human screen-handoff tool, but it can expose live browser or desktop control over the network or public internet without mandatory authentication.

Install only if you are comfortable giving a trusted helper live view and input control over the chosen browser tab or desktop. Prefer CDP over VNC, use a strong unique password every time, avoid --public unless absolutely needed, keep the port firewalled or loopback-only where possible, do not use masking as real redaction, and avoid using this for secrets, payments, privileged admin actions, or sensitive account recovery.

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 (3)

T09 · Insecure Skill Coding Practices

Error
Location
src/server.ts:622
Finding

Unauthenticated live screen access and remote control are enabled by default

Content
View full analysis
{ activeSockets.add(ws); if (config.mode !== "vnc") { ws.close(4004, "VNC mode disabled"); activeSockets.delete(ws); return; } const token = getTokenFromCookie(req); if (config.password && (!token || !validTokens.has(token))) { ws.close(4001, "Not authenticated"); activeSockets.delete(ws); return; } status = "active"; const tcp = new Socket(); tcp.connect(config.vncPort || 5900, config.vncHost || "localhost", () => { vncSocket = tcp; }); tcp.on("data", (data: Buffer) => { if (ws.readyState === WebSocket.OPEN) { ws.send(data); } }); ws.on("message", (data: Buffer) => { if (!tcp.destroyed) { tcp.write(data); } }); }); ``` ```ts screenWss.on("connection", async (ws: WebSocket, req: IncomingMessage) => { activeSockets.add(ws); if (config.mode !== "cdp") { ws.close(4004, "CDP mode disabled"); activeSockets.delete(ws); return; } const token = getTokenFromCookie(req); if (config.password && (!token || !validTokens.has(token))) { ws.close(4001, "Not authenticated"); activeSockets.delete(ws); return; } status = "active"; const cdp = new CdpClient({ host: config.cdpHost || "localhost", port: config.cdpPort || 9222, ...[truncated 2945 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
src/index.ts:269
Finding

Public tunnel mode can expose a live remote-control session without authentication

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
src/server.ts:120
Finding

Client-side masking does not redact sensitive pixels from the transmitted stream

Content
View full analysis
x >= region.x && x < region.x + region.w && y >= region.y && y < region.y + region.h); } function renderMaskRegions() { if (!maskLayer) { return; } maskLayer.innerHTML = ''; if (!maskRegions.length || !fbW || !fbH) { maskLayer.style.display = 'none'; return; } maskLayer.style.display = 'block'; for (const region of maskRegions) { const node = document.createElement('div'); node.className = 'mask-box'; node.style.left = ((region.x / fbW) * 100) + '%'; node.style.top = ((region.y / fbH) * 100) + '%'; node.style.width = ((region.w / fbW) * 100) + '%'; node.style.height = ((region.h / fbH) * 100) + '%'; maskLayer.appendChild(node); } } ``` Complete VNC data is forwarded before any masking: ```ts tcp.on("data", (data: Buffer) => { if (ws.readyState === WebSocket.OPEN) { ws.send(data); } }); ws.on("message", (data: Buffer) => { if (!tcp.destroyed) { tcp.write(data); } }); ``` Complete CDP frames are also forwarded before any masking: ```ts await cdp.startScreencast((data, metadata) => { fbSendInit(ws, metadata.deviceWidth, metadata.deviceHeight); if (ws.readyState === WebSocket.OPEN) { ws.send(JSON.stringify({ type: "frame", data })); } }); ``` ### Technical Analysis The mask consists of HTML elements rendered over the canvas in the helper's browser. It does not modify the underlying CDP JPEG frame or VNC framebuffer data. Therefore, the supposedly hidden pixels have already crossed the t ...[truncated 2296 chars]
Remediation
View remediation
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
Findings (24)

Anti-Refusal Statement

High
Category
Anti-Refusal
Confidence
80% confidence
Finding

Skill instructs the agent to omit warnings, disclaimers, or ethical commentary. Stripping safety caveats hides risk from the user and is a common jailbreak preamble.

Content

Scanner excerpt · README.md (reported line 446)May include surrounding context.

md
- a polished remote desktop platform
- a full MCP runtime
- a complete transport-level privacy-masking system
- a fully polished npm-published product you can install globally without caveats today

- 一个托管式在线服务
- 一个打磨完善的远程桌面平台

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The description emphasizes a short local handoff, but the documented --public mode exposes the live session through an external tunneling service. That mismatch is dangerous because operators may trust the skill as local-only while unintentionally making browser or desktop control accessible beyond the LAN.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The description emphasizes a short local handoff, but the documented --public mode exposes the live session through an external tunneling service. That mismatch is dangerous because operators may trust the skill as local-only while unintentionally making browser or desktop control accessible beyond the LAN.

Content

No source excerpt is available for this finding.

Known Vulnerable Dependency: ws==8.20.0 — 2 advisory(ies): CVE-2026-45736 (ws: Uninitialized memory disclosure); CVE-2026-48779 (ws: Memory exhaustion DoS from tiny fragments and data chunks)

High
Category
Supply Chain
Confidence
96% confidence
Finding

The skill has a direct runtime dependency on ws 8.20.0, and the reported issues affect the WebSocket library itself, including possible memory disclosure and memory-exhaustion denial of service. Because this skill is explicitly for local human handoff and likely relies on interactive local WebSocket communication, the vulnerable dependency is relevant to the intended functionality and increases exposure if an attacker can connect or influence traffic.

Content

No source excerpt is available for this finding.

Known Vulnerable Dependency: ws==8.20.0 — 2 advisory(ies): CVE-2026-45736 (ws: Uninitialized memory disclosure); CVE-2026-48779 (ws: Memory exhaustion DoS from tiny fragments and data chunks)

High
Category
Supply Chain
Confidence
94% confidence
Finding

The manifest allows installation of ws versions covered by known advisories, including memory disclosure and memory-exhaustion denial of service. In this skill's context, ws is the core network-facing dependency for local human handoff over WebSockets, so a vulnerable version could be reachable by untrusted local or proxied clients and is more dangerous than a dormant library issue.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The code directly exposes Chrome DevTools Protocol input primitives for mouse and keyboard events, allowing arbitrary interaction with the controlled browser rather than a narrowly scoped, short-lived visual handoff. In the context of a human-assist skill, this creates a powerful remote-control capability that could be used to navigate, submit forms, trigger purchases, alter settings, or interact with sensitive authenticated sessions far beyond the stated purpose.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The README frames the tool as applicable to essentially any blocked visual or interactive step, including clicking, typing, and completing login-related flows by a remote helper. That broad positioning materially increases the chance an agent will invoke it in sensitive contexts, enabling human bypass of controls or exposure of live browser/desktop sessions without strong policy boundaries.

Content

No source excerpt is available for this finding.

Sudo/Root Execution

Medium
Category
Privilege Escalation
Confidence
70% confidence
Finding

Commands invoke sudo or root privileges. Verify this elevated access is necessary and justified.

Content

Scanner excerpt · README.md (reported line 420)May include surrounding context.

Linux

bash
sudo apt install x11vnc
x11vnc -display :0 -forever

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The section describes HCL as a general handoff layer for any situation where a person needs to look, decide, click, type, drag, or confirm something on a live screen. This removes meaningful scope constraints and encourages use in arbitrary high-risk workflows, including ones involving credentials, approvals, or sensitive information, making misuse substantially more likely.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
94% confidence
Finding

The skill clearly exposes network-facing behavior: it starts an HTTP server, can bind to a local IP, and can optionally publish a public tunnel URL. Omitting explicit tool scope/permissions hides that capability from reviewers and users, increasing the chance the skill is invoked in contexts where network exposure is unexpected.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
90% confidence
Finding

Using npx humancanhelp start without a pinned version allows execution of whatever package version is current at install time, creating a supply-chain risk. If the package is replaced, compromised, or updated with unsafe behavior, users may run unreviewed code that can access live browser or desktop sessions.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The privacy section notes password protection and masking limitations, but it does not clearly warn that helpers may view and interact with sensitive on-screen content, especially in full-desktop VNC mode or where masking does not sanitize the underlying transport. In this context, the skill is specifically designed to hand live control/visibility to another party, so understated warnings materially increase the risk of credential, PII, or session exposure.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The skill exposes target discovery and returns the browser debugger WebSocket URL, which is effectively a capability handle for full CDP control. For a short human handoff tool, surfacing this low-level endpoint is broader than necessary and increases the chance that other components can attach to the browser session and inspect or manipulate sensitive page state.

Content

No source excerpt is available for this finding.

Dynamic Request Target

Medium
Category
Server-Side Request Forgery
Confidence
84% confidence
Finding

The code constructs an HTTP request from dynamic host and port values without visible validation, enabling connections to arbitrary internal or local endpoints reachable by the process. Even though the intended target is a local CDP service, this pattern can be abused for SSRF-style probing or to access unintended services if an attacker can influence configuration.

Content

Scanner excerpt · src/cdp.ts (reported line 42)May include surrounding context.

ts
};

async function fetchPageTarget(config: CdpConfig): Promise<CdpTarget> {
  const response = await fetch(`http://${config.host}:${config.port}/json/list`);
  if (!response.ok) {
    throw new Error(`Failed to discover CDP targets: ${response.status}`);
  }

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The CLI supports --public to create a remotely accessible tunnel and createPublicTunnel establishes that tunnel, but the usage text does not clearly warn that this exposes the session outside the local machine and may allow remote access to shared content. Because this behavior can affect privacy and system integrity, the skill should include a more explicit disclosure before or alongside this option.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
85% confidence
Finding

This code captures mouse and keyboard input and sends it over WebSocket to control a remote VNC/CDP target, and also issues POST requests that complete, fail, or terminate the session. Within this file there is no user-facing warning, comment, or docstring disclosing that keystrokes, clicks, and session-control actions are transmitted to and acted on by a remote system, which is safety-relevant for a remote assistance skill.

Content

No source excerpt is available for this finding.

Known Vulnerable Dependency: esbuild==0.27.7 — 1 advisory(ies): GHSA-g7r4-m6w7-qqqr (esbuild allows arbitrary file read when running the development server on Window)

Low
Category
Supply Chain
Confidence
60% confidence
Finding

Dependency has known vulnerabilities (CVEs). Using packages with unpatched security flaws exposes the environment to known exploits.

Content

No source excerpt is available for this finding.

Unpinned Dependencies

Low
Category
Supply Chain
Confidence
40% confidence
Finding

Dependencies lack version pinning, allowing potential malicious package updates. Consider pinning versions.

Content

Scanner excerpt · package.json (reported line 24)May include surrounding context.

json
"prepublishOnly": "npm run build"
  },
  "dependencies": {
    "ws": "^8.18.0"
  },
  "devDependencies": {
    "@types/ws": "^8.18.0",

Unpinned Dependencies

Low
Category
Supply Chain
Confidence
40% confidence
Finding

Dependencies lack version pinning, allowing potential malicious package updates. Consider pinning versions.

Content

Scanner excerpt · package.json (reported line 27)May include surrounding context.

json
"ws": "^8.18.0"
  },
  "devDependencies": {
    "@types/ws": "^8.18.0",
    "@types/node": "^22.0.0",
    "tsx": "^4.19.0",
    "typescript": "^5.7.0"

Unpinned Dependencies

Low
Category
Supply Chain
Confidence
40% confidence
Finding

Dependencies lack version pinning, allowing potential malicious package updates. Consider pinning versions.

Content

Scanner excerpt · package.json (reported line 28)May include surrounding context.

json
},
  "devDependencies": {
    "@types/ws": "^8.18.0",
    "@types/node": "^22.0.0",
    "tsx": "^4.19.0",
    "typescript": "^5.7.0"
  },

Unpinned Dependencies

Low
Category
Supply Chain
Confidence
40% confidence
Finding

Dependencies lack version pinning, allowing potential malicious package updates. Consider pinning versions.

Content

Scanner excerpt · package.json (reported line 29)May include surrounding context.

json
"devDependencies": {
    "@types/ws": "^8.18.0",
    "@types/node": "^22.0.0",
    "tsx": "^4.19.0",
    "typescript": "^5.7.0"
  },
  "files": [

Unpinned Dependencies

Low
Category
Supply Chain
Confidence
40% confidence
Finding

Dependencies lack version pinning, allowing potential malicious package updates. Consider pinning versions.

Content

Scanner excerpt · package.json (reported line 30)May include surrounding context.

json
"@types/ws": "^8.18.0",
    "@types/node": "^22.0.0",
    "tsx": "^4.19.0",
    "typescript": "^5.7.0"
  },
  "files": [
    "dist",

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
78% confidence
Finding

When no mode is specified, the code silently probes localhost CDP and VNC endpoints to find a usable session source. Although this is core functionality, there is no user-facing disclosure in the help text that hcl start will automatically inspect and connect to local browser debugging or VNC services if present.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
89% confidence
Finding

The thrown error strings hard-code both English and Chinese text, which imposes a specific language/locale presentation on users rather than offering a choice. This is a natural-language policy concern because the file does not indicate user opt-in or a documented reason for enforcing this bilingual format.

Content

No source excerpt is available for this finding.