The Agent Skills Directory

[PROMPT_INJECTION]: Analysis identified a surface for indirect prompt injection, a common characteristic of tools that aggregate and display user-provided content back to the AI agent.
Ingestion points: Food names provided by the user are stored in the calorie_data.db file via the add_entry function in scripts/calorie_tracker.py.
Boundary markers: Data is presented to the agent in a text-based table (using | and = symbols) in the summary and list_entries functions, which may not sufficiently isolate user content from the agent's internal instruction stream.
Capability inventory: The script's operations are confined to the local workspace; it manages an SQLite database and prints to standard output. It does not have network access, sensitive file system permissions, or the ability to execute arbitrary system commands.
Sanitization: The skill does not perform sanitization, escaping, or filtering of user-inputted food names to prevent them from containing malicious instructions.

calorie-counter