Security audit

NWO Robotics

Security checks for vulnerabilities and agentic risk

Overview

This skill appears intended to control robots, but it gives broad natural-language commands high-impact physical authority without adequate scoping or safety controls.

Review before installing. Use only with authorized robot/IoT accounts in a controlled environment, restrict API key permissions, keep NWO_API_URL pinned to the intended HTTPS service, and do not use it for unattended or production physical operations unless confirmation, command scoping, and safety interlocks are added.

A.I.G

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 (4)

T09 · Insecure Skill Coding Practices

Error

Location: index.js:7
Finding: Configurable API Base URL Can Exfiltrate Credentials and Operational Data
Content: View full analysis
Remediation: View remediation

T09 · Insecure Skill Coding Practices

Error

Location: index.js:13
Finding: Declared Command Allowlist Is Not Enforced and Unknown Instructions Are Executed as General Tasks
Content: View full analysis
Remediation: View remediation

T09 · Insecure Skill Coding Practices

Error

Location: index.js:99
Finding: Substring-Based Intent Routing Can Trigger Unintended Physical and Safety Operations
Content: View full analysis
Remediation: View remediation

T09 · Insecure Skill Coding Practices

Warning

Location: index.js:73
Finding: Untrusted API Response Content Is Returned Without Output Validation
Content: View full analysis
({})); throw new Error(error.message || `API error: ${response.status}`); } return await response.json(); ``` ```js const statusList = data.agents.map(agent => `• ${agent.name || agent.id}: ${agent.status || 'unknown'}` ).join('\n'); return `Robot Status:\n${statusList}`; ``` ```js const objects = data.objects.map(obj => `• ${obj.label} (${obj.confidence}% confidence) at [${obj.x}, ${obj.y}]` ).join('\n'); return `Detected ${data.objects.length} objects:\n${objects}`; ``` ```js const readings = data.sensors.map(sensor => `• ${sensor.location || sensor.id}: ${sensor.value}${sensor.unit || ''}` ).join('\n'); return `Sensor Readings:\n${readings}`; ``` ```js return `Task submitted. ${data.message || 'Processing...'}`; } catch (error) { return `Error processing task: ${error.message}`; } ``` ### Technical Analysis Multiple fields received from the remote API are interpolated directly into skill output without schema validation, type checks, length limits, control-character filtering, or trust-boundary labeling. This includes error messages, agent names and statuses, object labels, sensor data, and general-task messages. The user-input sanitizer does not mitigate this issue because it applies only to incoming user text, not API responses. If the remote API is compromised, if device-controlled metadata is attacker-influenced, or if requests are redirected through the configurable API base URL, an attacker can return arbitrary text that the skill presents as ordinary output. When skill output is ...[truncated 1414 chars]
Remediation: View remediation

SkillSpector

Vulnerability Patterns

Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration

Findings (12)

Missing User Warnings

High

Category: Not specified by scanner
Confidence: 94% confidence
Finding: The skill documentation includes safety-critical commands such as 'Stop all robots', movement, patrol mode, and manipulation tasks without any visible warnings, prerequisites, or operator confirmation guidance. In the context of robot and IoT control, missing warnings increases the chance of accidental invocation, unsafe operation near people/equipment, and misuse by operators who may not understand the physical consequences.
Content: No source excerpt is available for this finding.

Vague Triggers

High

Category: Not specified by scanner
Confidence: 97% confidence
Finding: The README explicitly advertises 'Any natural language instruction' as a supported general task, which implies extremely broad activation and weak scope constraints. In a robotics skill, this is dangerous because ambiguous or unrelated user phrasing could trigger physical-world actions or cause the agent to over-delegate unsafe commands to the backend.
Content: No source excerpt is available for this finding.

Missing User Warnings

High

Category: Not specified by scanner
Confidence: 96% confidence
Finding: The handler routes unmatched or loosely matched user input into autonomous execution paths without an explicit warning, review step, or confirmation. Because this skill controls robots, free-form remote task planning can translate ambiguous or malicious prompts into unsafe physical actions, making the context substantially more dangerous than a typical software-only agent.
Content: No source excerpt is available for this finding.

Missing User Warnings

High

Category: Not specified by scanner
Confidence: 97% confidence
Finding: This skill enables control of robots and IoT devices, including examples like movement and emergency stop, but the manifest provides no user-facing warning that commands may trigger real-world, safety-critical physical actions. Without explicit warnings and operator safeguards, users or upstream agents may invoke dangerous actions in inappropriate contexts, potentially causing injury, equipment damage, or operational disruption.
Content: No source excerpt is available for this finding.

Missing User Warnings

Medium

Category: Not specified by scanner
Confidence: 94% confidence
Finding: The skill explicitly enables natural-language control of robots and IoT systems in the physical world, but the description does not prominently warn that user prompts can trigger movement, task execution, or emergency-stop behavior with real-world safety consequences. In a robotics context, omission of this warning increases the chance of accidental or uninformed activation of hazardous actions by users or downstream agents.
Content: No source excerpt is available for this finding.

Missing User Warnings

Medium

Category: Not specified by scanner
Confidence: 95% confidence
Finding: The example commands include safety-critical or potentially disruptive actions such as "Stop all robots," "Pick up the red box," and "Patrol mode on" without any caution about operational impact, authorization, or environment safety checks. Providing these as normal examples can normalize dangerous use and lead users or agents to invoke physical actions without understanding risks to people, equipment, or facilities.
Content: No source excerpt is available for this finding.

Missing User Warnings

Medium

Category: Not specified by scanner
Confidence: 92% confidence
Finding: The emergency-stop examples normalize issuing safety-critical commands without any warning, authorization note, or operational context. In a robotics skill, this can encourage users or downstream agents to trigger halts in live environments, potentially causing safety incidents, workflow disruption, or unsafe robot states depending on the machinery and process.
Content: No source excerpt is available for this finding.

Missing User Warnings

Medium

Category: Not specified by scanner
Confidence: 89% confidence
Finding: The movement-command examples show direct robot navigation instructions without warnings about collision risk, workspace clearance, human presence, or need for validation before execution. In a skill that controls physical robots, this omission can make unsafe movement requests appear routine and low-risk, increasing the chance of accidental equipment damage or harm in real-world settings.
Content: No source excerpt is available for this finding.

Intent-Code Divergence

Medium

Category: Not specified by scanner
Confidence: 92% confidence
Finding: The skill advertises 'secure robot control' but accepts natural-language input and passes it to broad remote execution endpoints such as plan_and_execute and execute_task. In a robotics context, this mismatch is dangerous because users may assume constrained behavior while the implementation permits open-ended actions that can affect physical systems.
Content: No source excerpt is available for this finding.

Intent-Code Divergence

Medium

Category: Not specified by scanner
Confidence: 98% confidence
Finding: The code defines an ALLOWED_COMMANDS allowlist but never enforces it. As a result, any unrecognized input falls through to handleGeneralTask(), which forwards arbitrary user instructions to a remote plan_and_execute endpoint, enabling broad robot task execution despite the stated validation controls.
Content: No source excerpt is available for this finding.

Missing User Warnings

Medium

Category: Not specified by scanner
Confidence: 81% confidence
Finding: The function immediately performs a remote emergency_stop call affecting all robots. While the return message confirms the action afterward, there is no pre-execution warning or confirmation prompt informing the user that this command will remotely halt all robots.
Content: No source excerpt is available for this finding.

Vague Triggers

Medium

Category: Not specified by scanner
Confidence: 92% confidence
Finding: The manifest advertises broad natural-language control of robots and IoT devices without narrowing allowed actions, environments, or approval boundaries. In a safety-relevant domain, this can cause overbroad invocation and execution of hazardous commands from ambiguous prompts, increasing the chance of unintended physical actions.
Content: No source excerpt is available for this finding.

Static analysis

Detected: suspicious.env_credential_access

Environment variable access combined with network send.

Critical

Code: suspicious.env_credential_access
Location: index.js:7