The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection by ingesting and processing untrusted data from external sources such as food labels, restaurant menus, and user-provided meal photos.
Ingestion points: The skill reads meal descriptions, photos, and nutrition labels in estimation.md and labels.md. It also processes external restaurant nutrition pages as described in restaurants.md.
Boundary markers: There are no explicit instructions to the agent to ignore or delimit potentially malicious text embedded within the nutrition labels or text logs.
Capability inventory: The skill is capable of reading and writing to specific local directories (~/Clawic/data/calories/) to maintain the food library and user logs. It does not possess capabilities for arbitrary command execution or unauthorized network communication.
Sanitization: While the instructions emphasize extracting numeric and specific textual data, there is no technical sanitization or escaping of the ingested content before it enters the agent's context.
[SAFE]: The skill employs localized file storage for user data and configuration within the ~/Clawic/ directory, adhering to standard data management practices for agent skills.
[SAFE]: Detailed medical guardrails and 'Red Flags' are implemented in safety.md and SKILL.md to protect users, demonstrating a focus on safety and ethical boundaries.
[SAFE]: All referenced external links point to the vendor's official domain (clawic.com) or established eating disorder helplines, which are trusted resources and do not represent a security risk.

Calorie Tracker