The Agent Skills Directory

[COMMAND_EXECUTION]: The shell script scripts/save-plan.sh is vulnerable to SQL injection by using unsanitized shell arguments in a database command.
The variables $1, $2, and $3 are interpolated directly into the sqlite3 INSERT statement, allowing malicious input to break out of the query.
[COMMAND_EXECUTION]: The /meals pref command instruction in SKILL.md contains a SQL injection vulnerability.
It encourages the agent to insert <key> and <value> strings directly into a sqlite3 query, which can be exploited if the inputs contain single quotes or SQL control characters.
[PROMPT_INJECTION]: The skill has an indirect prompt injection surface because it processes untrusted data from a local database without boundaries or sanitization.
Ingestion points: Data is read from the fridge table in /data/workspace/pantry.db as shown in the /meals plan command logic in SKILL.md.
Boundary markers: No delimiters or instructions to ignore embedded commands are present in the prompt interpolation logic.
Capability inventory: The skill can execute bash scripts and perform sqlite3 operations on the filesystem.
Sanitization: No sanitization, escaping, or validation of the data retrieved from the database is performed before it is used to generate the meal plan.

meal-planner