The Agent Skills Directory

[PROMPT_INJECTION]: The skill is configured to ingest external data from various JSON files (e.g., nutrition logs, fitness trackers). This represents a potential surface for indirect prompt injection if those files are attacker-controlled. However, the agent's capabilities are limited to basic file operations (Read, Grep, Glob, Write), and it lacks the ability to execute system commands or access the network, mitigating the risk of exploitation.
[DATA_EXFILTRATION]: The skill accesses sensitive personal files including profile.json (containing BMI/weight) and health-specific trackers for hypertension and diabetes. This data exposure is intrinsic to the skill's primary function of health analysis. There are no mechanisms detected (such as HTTP requests or external shell commands) that would allow this data to be exfiltrated from the local environment.

nutrition-analyzer