Security audit
Security checks for vulnerabilities and agentic risk
This food-tracking skill is a plain Markdown workflow that stores user-submitted food data locally and discloses that persistence, with privacy considerations users should understand before use.
Install this only if you want a local food history and preference database. Review ~/food/memory.md periodically and delete entries you do not want retained, especially allergies, restrictions, eating patterns, restaurants, and recipes tied to personal routines.
Hidden instructions were detected in comments or invisible text. These could contain malicious directives. Manual review is recommended.
Format:
### Preferences
<!-- Their food preferences and restrictions. Format: "item: type" -->
<!-- Examples: nuts: allergy, gluten: intolerance, vegetarian: choice -->
### Products
The skill advertises that it will absorb 'ANY food input' and auto-classify it, which is overly broad and can cause the agent to ingest unintended sensitive content such as photos, labels, and freeform text without clear scope boundaries. In a food-tracking context this increases the chance of collecting personal, behavioral, or health-adjacent data beyond what the user reasonably expects.
The skill explicitly persists dietary preferences, restrictions, products, places, recipes, and eating patterns in a local memory file, but it does not present a clear privacy notice or obtain explicit consent for long-term retention. Because this data can reveal sensitive health, religious, lifestyle, and behavioral information, silent persistence materially raises privacy and profiling risk.
The auto-classification rules rely on broad, weak cues such as generic phrases like "I ate X" or any photo with certain characteristics, which can misroute user input into logging, storage, or recommendation flows without sufficient confirmation. In a food-tracking skill, this increases the chance of unintended collection, incorrect health records, and actions taken on sensitive dietary data based on misclassification.
The workflow states that allergy or avoidance preferences should be stored permanently, but gives no user notice, consent step, or control over retention. Because dietary restrictions can reveal sensitive health information or religious practices, silent permanent persistence creates a meaningful privacy and compliance risk.
These pipelines repeatedly instruct the system to store meal, product, recipe, and linked inventory data without any privacy notice, retention limit, or user control. In aggregate, this can create a detailed behavioral profile of eating habits, shopping patterns, routines, and possible health-related inferences, making the skill context more sensitive rather than less.
No suspicious patterns detected.