Security audit

The Spatiotemporal Rendering Engine

Security checks for vulnerabilities and agentic risk

Overview

This smart-home timeline skill is not deceptive, but it should be reviewed because it saves unvalidated LLM-generated device schedules that may later drive real-world automation.

Review this before installing in any real smart-home environment. It is most appropriate for simulation or a sandbox unless a separate executor independently validates every device, action, value, and schedule and requires confirmation for sensitive actions such as heaters, locks, high-volume audio, robots, or medication workflows. Treat any microphone, mmWave, behavioral, or elder-care use as privacy-sensitive and require explicit consent and clear retention controls.

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 (1)

T09 · Insecure Skill Coding Practices

Warning

Location: skill.py:48
Finding: Unvalidated LLM Output Persisted as Actionable IoT Timeline Instructions
Content: View full analysis
{kf.get('device_id')}") print(f" Action: {kf.get('rendered_state')} ({kf.get('reason_nlp')})") tracks_db = load_json(TRACKS_FILE) if "scheduled_tracks" not in tracks_db: tracks_db["scheduled_tracks"] = [] tracks_db["scheduled_tracks"].append(track_result) save_json(TRACKS_FILE, tracks_db) ``` ### Technical Analysis The skill embeds user-controlled natural-language input and hardware-mount metadata into a single LLM prompt. Both sources must be treated as untrusted because the interactive user controls the intent, while local processes o ...[truncated 2911 chars]
Remediation: View remediation

SkillSpector

Vulnerability Patterns

Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
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

Findings (7)

Ssd 1

High

Category: Not specified by scanner
Confidence: 96% confidence
Finding: User-controlled natural language is inserted directly into the orchestration prompt that determines device actions, so an attacker can craft input that semantically overrides constraints or manipulates the model into generating unsafe or unauthorized keyframes. In this skill context, the generated output affects IoT automation timelines, making prompt injection more dangerous because it can influence real-world device behavior rather than only text output.
Content: No source excerpt is available for this finding.

Vague Triggers

Medium

Category: Not specified by scanner
Confidence: 95% confidence
Finding: The skill uses a broad spoken trigger phrase ('Make a wish!') tied to immediate environmental actions, which can be activated accidentally, by guests, or by media playback. In a smart-home context, ambiguous voice triggers increase the chance of unintended automation and can be abused to manipulate lighting/audio behavior without clear user intent.
Content: No source excerpt is available for this finding.

Missing User Warnings

Medium

Category: Not specified by scanner
Confidence: 96% confidence
Finding: The skill describes continuous or event-driven microphone monitoring to detect natural-language phrases but does not present any explicit privacy notice, consent model, or data-handling limitation. Audio monitoring inside a home is highly privacy-sensitive because it may capture bystanders, private conversations, and non-user speech without meaningful awareness.
Content: No source excerpt is available for this finding.

Missing User Warnings

Medium

Category: Not specified by scanner
Confidence: 98% confidence
Finding: The skill describes collecting and correlating intimate in-home behavioral data, including breathing patterns, posture, emotional distress, pet vocalizations, long-term movement history, and elder activity status, without clear disclosure or safeguards. This level of behavioral inference can reveal health, emotional, and daily-routine information, making misuse or overcollection especially invasive and potentially harmful.
Content: No source excerpt is available for this finding.

Missing User Warnings

Medium

Category: Not specified by scanner
Confidence: 93% confidence
Finding: The skill describes autonomous physical actions—dispatching a robot, transporting medication and water, and issuing reminders—without clearly warning that the system can affect the real world. Real-world actuation creates safety and reliability risks if the inference is wrong, conditions change, or the robot interferes with people, pets, or ongoing activities.
Content: No source excerpt is available for this finding.

Missing User Warnings

Medium

Category: Not specified by scanner
Confidence: 89% confidence
Finding: The skill sends user intent and device inventory to an HTTP API endpoint without notifying the user or obtaining consent. Even though the endpoint is localhost, it still exposes sensitive behavioral data and local device metadata to another service, which may log, forward, or be replaced by an untrusted local process.
Content: No source excerpt is available for this finding.

Intent-Code Divergence

Low

Category: Not specified by scanner
Confidence: 94% confidence
Finding: The message at L149 says the rendered track was 'securely injected into Timeline DB', which implies a secured database operation. In reality, L143-L147 simply load a local JSON file, append data, and write it back to disk, so the wording actively overstates and mischaracterizes what the code does.
Content: No source excerpt is available for this finding.

Static analysis

No suspicious patterns detected.