The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to execute shell commands via docker compose exec. It takes user-provided SQL queries and interpolates them directly into a shell command string, which can lead to command injection if the input is not strictly validated by the agent.\n- [CREDENTIALS_UNSAFE]: The instructions require the agent to read database passwords from CLAUDE.local.md and pass them using the -p<PASSWORD> flag in the command line. This practice is insecure as it exposes sensitive credentials in the system's process list in plain text.\n- [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by ingesting and displaying data from database tables like wp_simple_history.\n
Ingestion points: SQL query results from WordPress database tables (SKILL.md).\n
Boundary markers: Absent; the skill does not provide instructions to treat the retrieved database content as untrusted or to ignore embedded instructions.\n
Capability inventory: Shell execution via docker compose exec and file system access (SKILL.md).\n
Sanitization: Absent; query results are rendered directly to the agent's context without escaping or validation.

sql