Back to skill

Security audit

S2-SP-OS Wave/Spectrum Radar

Security checks for vulnerabilities and agentic risk

Overview

This skill needs review before installation because it presents itself as passive hardware sensing, but the packaged code is broken or mocked and the examples encourage autonomous smart-home and audio follow-up actions under a weak persistent consent flag.

Review this skill carefully before installing. Do not rely on its output for real occupancy or health decisions until it has a runnable implementation that actually reads and validates the intended hardware. Remove or constrain the autonomous lighting, audio-monitoring, and emergency examples; require explicit per-action user approval; avoid persisting the consent flag in startup files; and pin dependencies in a controlled environment.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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
Findings (4)

T09 · Insecure Skill Coding Practices

Error
Location
spectrum.py:26
Finding

Fabricated sensor values are represented as genuine hardware telemetry

Content
View full analysis
dict: return { "occupancy": True, "motion_state": "Active", "distance_m": "Unknown_GPIO_Limitation" } def read_uart_telemetry(self) -> dict: if not self.port: return {"error": "UART mode requires a --port argument (e.g., /dev/ttyUSB0)"} # Simulated parsed data from MCU internal_distance_m = 1.45 internal_heart_bpm = 58 internal_breath_bpm = 15 heart_status, breath_status = self._quantize_health_data( internal_heart_bpm, internal_breath_bpm ) # Ephemeral privacy: Destroy exact numbers del internal_heart_bpm del internal_breath_bpm return { "occupancy": True, "motion_state": "Static_MicroMotion", "distance_m": internal_distance_m, "heart_status": heart_status, "breathing_status": breath_status } ``` ```python if args.mode == "gpio": core_tensors = client.read_gpio_state() vendor_nl = "Hardware GPIO parsed. Pure passive occupancy established. / 硬件GPIO直读完毕,确立纯粹的被动占位状态。" elif args.mode == "uart": result = client.read_uart_telemetry() if "error" in result: print(json.dumps(result, ensure_ascii=False)) sys.exit(1) core_tensors = result vendor_nl = "UART Telemetry parsed. Biometric data successfully QUANTIZED for passive observation. / 串口遥测帧解析完毕,生物数据已量化,仅供被动观察。" ``` ### Technical Analysis The implementation does not perform any GPIO or UART operation. There is no GPIO-library call, GPIO pin read, `serial.Serial` construction, serial-port opening, frame read, frame validation, or hardware error handling. Instead, GPIO mode always reports an occupied and active space. UART mode only checks wh ...[truncated 1559 chars]
Remediation
View remediation

T01 · Skill Instruction Hijacking

Error
Location
AGENT-EXAMPLES.md:17
Finding

Agent examples direct autonomous cross-device actuation and surveillance

Content
View full analysis
Remediation
View remediation

T08 · Insecure Dependencies

Warning
Location
edge-setup-guide.md:16
Finding

Unpinned third-party dependency creates a mutable supply-chain boundary

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
edge-setup-guide.md:21
Finding

Persistent environment variable turns consent into blanket authorization

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • 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 (8)

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The skill claims to perform passive mmWave sensing over GPIO/UART with privacy-gated biometric quantization, but the documented behavior appears inconsistent with any real hardware interaction. This is dangerous because agents or operators may rely on fabricated or simulated occupancy/biometric outputs as if they were trustworthy sensor data, which can lead to unsafe downstream decisions and privacy misrepresentations.

Content

No source excerpt is available for this finding.

Lp1

High
Category
MCP Least Privilege
Confidence
75% confidence
Finding

The skill uses 'env' capability that is not listed in its permissions. This may indicate deceptive intent or missing permission declarations.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The examples explicitly model the agent taking autonomous follow-on actions based on sensor output, despite the documentation claiming the skill is purely passive. This creates instruction-channel ambiguity: downstream agents may treat these examples as normative guidance and trigger lighting or other actuators without explicit user approval or a separate policy gate.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The emergency example goes beyond passive sensing and instructs the agent to verify anomalies via additional monitoring and escalate to emergency routines. In a safety- and privacy-sensitive context, this can cause an agent to initiate surveillance or emergency actions from noisy sensor data, increasing the risk of privacy intrusion, false alarms, and unauthorized intervention.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The documentation normalizes sensitive audio monitoring and emergency escalation without clearly warning users that these actions may collect additional intimate data or trigger consequential real-world responses. Because the skill concerns biometrics and home occupancy, omission of consent and warning language makes the example more dangerous than a generic automation case.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
78% confidence
Finding

The file title and scenario headings combine English and Chinese text, implicitly imposing a mixed-language experience without stating that language selection is optional or user-configurable. Under the policy, language constraints should be opt-in or clearly justified.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
78% confidence
Finding

The file presents core instructions and descriptions in both English and Chinese, but it does not state whether language selection is user-driven or required by policy. This can violate language/locale policy expectations if the organization requires explicit user opt-in for language behavior.

Content

No source excerpt is available for this finding.

Vague Triggers

Low
Category
Not specified by scanner
Confidence
82% confidence
Finding

The invocation guidance tells an agent to execute hardware-facing commands but does not define when the skill should or should not be used, what user authorization is required beyond an environment variable, or what contexts are forbidden. Without clear trigger boundaries, an agent could invoke the skill in inappropriate environments or on unintended devices, increasing the chance of privacy-invasive or unsafe sensing attempts.

Content

No source excerpt is available for this finding.