The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/db-init.sh utility script is vulnerable to code injection because it interpolates shell variables $DB_FILE and $SCHEMA directly into a Python heredoc block. Since the shell expands these variables before passing the content to the Python interpreter, an attacker who can influence these parameters can execute arbitrary Python code. This represents a significant risk if the agent initializes databases based on user-provided names or configuration.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its handling of untrusted data and its high-privilege SQL capabilities. 1. Ingestion points: The skill ingests untrusted data through functions like set_key, insert_metric, and enqueue, as well as via SQL parameters. 2. Boundary markers: Absent. The skill documentation and implementation do not include instructions for the agent to use delimiters or ignore embedded commands within stored data. 3. Capability inventory: The skill provides tools for raw SQL execution (query, execute) and database initialization via shell scripts. 4. Sanitization: Absent. While parameterized query examples are shown, the skill provides raw access to the database engine and the initialization utility lacks argument validation.

sqlite-quick