Back to skill

Security audit

S2-SWM Boundary Scanner Engine(动态边界扫描器)

Security checks for vulnerabilities and agentic risk

Overview

The skill is a robotics-style boundary scanner, but it returns fixed simulated obstacle data as successful radar scans, which could mislead navigation or safety decisions.

Review this carefully before installing in any live robotics, mobility, or safety workflow. It appears acceptable only as a demo or conceptual prototype unless it is changed to use real validated sensor input, clearly label simulation mode, fail closed when hardware is unavailable, and add privacy and human-control safeguards.

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

Error
Location
handler.py:43
Finding

Simulated Sensor Readings Are Presented as Successful Physical Boundary Scans

Content
View full analysis
dict: """ Simulates low-level mmWave radar point-cloud data. In a real project, this would connect to a radar serial port or ROS topic. """ # Assume that the embodied robot is walking toward a glass wall, # with a concrete pillar on its right. raw_sensor_data = { "Grid_Front": {"distance_m": 0.8, "rcs": 25.5}, "Grid_FrontRight": {"distance_m": 1.2, "rcs": 8.0}, "Grid_Right": {"distance_m": 3.0, "rcs": 0.5}, "Grid_BackRight": {"distance_m": 3.0, "rcs": 0.5}, "Grid_Back": {"distance_m": 3.0, "rcs": 0.5}, "Grid_BackLeft": {"distance_m": 3.0, "rcs": 0.5}, "Grid_Left": {"distance_m": 2.5, "rcs": 2.0}, "Grid_FrontLeft": {"distance_m": 1.8, "rcs": 25.5} } return raw_sensor_data def execute_boundary_scan(self, center_hex_code: str, heading_vector: str, step_size_mm: int) -> dict: """ Core action: perform a step-wise boundary scan. """ logging.info(f"Physical displacement {step_size_mm}mm; starting 9-grid mmWave boundary scan. Heading: {heading_vector}") # Obtain low-level radar readings sensor_data = self._simulate_mmwave_readings(heading_vector) ``` ### Technical Analysis The purported boundary scan does not communicate with mmWave hardware, a serial device, a ROS topic, or caller-supplied sensor measurements. It always invokes `_simulate_mmwave_readings`, which returns the same hard-coded distances and Radar Cross Section values. Although `heading_vector` is passed into the simulation function, it is not used to transform or select the readings. Consequently, every position, heading, and environment receives the same apparent topology. The configured `MMWAVE_COM_POR ...[truncated 2340 chars]
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • System Prompt LeakageDirect Leakage, Indirect Extraction, Tool-Based Exfiltration
  • 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
Findings (6)

Direct Prompt Extraction

High
Category
System Prompt Leakage
Confidence
85% confidence
Finding

Skill contains instructions that could directly expose system prompts, internal rules, or hidden instructions to users or external parties.

Content

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

md
* **Intrusion Percentage:** How much of the adjacent SSSU is consumed by a physical boundary.
* **Material Inference:** Derived from Radar Cross Section (RCS). Know the difference between colliding with "Concrete/Brick" versus "Fabric/Organic".

## 4. Output Directives (语义化输出)
When reporting spatial status to humans or upstream LLM reasoning modules, translate the tensors into precise multimodal causal text.
* *Example Compliant Output:* "Step displacement 500mm complete. Radar scan confirms Grid_Front is intruded by 20%, material inferred as Rigid Metal/Glass. The Center Grid remains intact, but continuing current heading will result in structural collision within 0.8 meters."

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The heading presents directives in Chinese only, which imposes a specific language on the interaction without indicating user choice or opt-in. This can violate language/locale policy when the skill does not offer an alternative language or document a justified locale restriction.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The skill mandates execute_boundary_scan after any movement where step_size_mm > 0, which is an extremely broad trigger and can cause frequent or unintended tool invocations. In an embodied/robotics context, this can create unnecessary sensing actions, resource exhaustion, or unsafe coupling between minor motion events and mandatory scanning behavior.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

This is a real safety/security issue because the skill advertises real-time mmWave boundary scanning on movement, but the implementation uses fixed simulated readings regardless of the environment. In an embodied robotics context, downstream agents may trust the output for navigation or collision avoidance, causing unsafe movement decisions, environmental misclassification, or failure to detect obstacles entirely.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The document explicitly recommends continuous environmental scanning with mmWave radar and optional camera/LiDAR inputs, but provides no notice, consent, retention, or bystander-privacy guidance. In an embodied AI context, this can normalize covert collection of spatial and potentially sensitive environmental data inside homes, workplaces, or other private spaces.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The skill describes forwarding multimodal prediction outputs to upstream LLMs and downstream robotic systems without warning about data sharing, model exposure, or unsafe automated actuation. Because these outputs may contain detailed environmental inferences and collision/causal predictions, ungoverned export can leak sensitive context and drive downstream physical actions without clear safety boundaries.

Content

No source excerpt is available for this finding.