The Agent Skills Directory

[COMMAND_EXECUTION]: The hook script scripts/block-unverified-destructive-db.py uses subprocess.run to call docker exec <container> env. This is a core part of its defensive logic, used to resolve the actual database name from a container's environment to verify its safety. The implementation uses a list of arguments for subprocess.run, which effectively prevents shell injection attacks.
[SAFE]: The skill adopts a 'fail-closed' security philosophy. If the hook script is unable to parse a command, resolve a container target, or verify a database name, it blocks the command by default rather than allowing it through. This prevents bypasses via malformed input.
[SAFE]: The installation steps involve common administrative tasks for AI agent hooks, such as marking the local script as executable (chmod +x) and updating the agent's settings.json. These actions are restricted to the skill's own scripts and configuration, representing intended and safe functionality.

docker-db-guardrails