Calorie Lookup

• Outcome: Pass
• Latest audit: May 28, 2026 at 11:44 AM UTC
• Version:
  0.3.0

Overview

This nutrition lookup skill is coherent and purpose-aligned, with expected API lookups, optional model-based translation/photo recognition, and local caching disclosed across its artifacts.

Install only if you are comfortable providing Spoonacular or USDA API keys and having meal descriptions or food photos processed by the configured nutrition APIs and model/sub-agent workflow. Avoid uploading sensitive images, and adjust or clear CALORIE_SKILL_CACHE_DB if you do not want local meal-query history retained.

SkillSpector

By NVIDIA

Vulnerability Patterns

• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration
• MCP Tool Poisoning: Hidden Instructions, Unicode Deception, Parameter Description Injection
• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands
• Privilege Escalation: Excessive Permissions, Sudo/Root Execution, Credential Access
• Supply Chain: Unpinned Dependencies, External Script Fetching, Obfuscated Code

Findings (5)

Tp4

High

• Category: MCP Tool Poisoning
• Confidence: 81% confidence
• Finding: The file presents capabilities that do not cleanly match the documented implementation, including claiming image recognition without visible image-handling logic and omitting material behaviors like local SQLite caching and translation pathways. This kind of mismatch is dangerous because users and reviewers may make trust decisions based on incomplete or inaccurate documentation, which can hide data handling, persistence, or unexpected processing paths.

Description-Behavior Mismatch

Medium

• Confidence: 93% confidence
• Finding: The guide introduces mandatory Decomposer and image-recognizer sub-agents plus LLM-based translation/recognition that are not disclosed in the skill metadata. This is a supply-chain transparency issue: users and integrators may believe the skill only performs API nutrition lookup, while their inputs or images are actually routed to additional model-driven components, changing the data flow and trust boundary.

Natural-Language Policy Violations

Medium

• Confidence: 88% confidence
• Finding: Forcing all non-English input through an LLM decomposer/translation path without user consent can expose potentially sensitive dietary, health, or personal context to an additional processing component. It also removes user control over whether their language input is transformed or sent to third-party model services, which is a meaningful privacy and policy risk in a consumer nutrition skill.

Missing User Warnings

Medium

• Confidence: 95% confidence
• Finding: The README advertises photo-based recognition and third-party API-backed nutrition lookup, but it does not clearly disclose that user-provided food photos and descriptions may be transmitted to external providers. This creates a real privacy and informed-consent issue because users may unknowingly send personal or sensitive dietary data to Spoonacular, USDA, or multimodal model providers.

Missing User Warnings

Medium

• Confidence: 94% confidence
• Finding: The workflow routes user food photos to a multimodal sub-agent/LLM without any explicit notice, consent step, or data-handling disclosure. Images often contain sensitive incidental data (faces, locations, receipts, surroundings), so silent transmission to an LLM increases privacy risk and may violate user expectations or policy requirements.

VirusTotal

64/64 vendors flagged this skill as clean.