The Agent Skills Directory

[REMOTE_CODE_EXECUTION] (HIGH): The skill documentation explicitly instructs users to execute curl -sSL https://canifi.com/skills/zoom/install.sh | bash. Piping unverified remote scripts to a shell is a critical security risk as the server can deliver malicious code based on user-agent or timing.
[REMOTE_CODE_EXECUTION] (HIGH): A second instance of the pipe-to-bash pattern is found for the setup script: curl -sSL https://canifi.com/install.sh | bash.
[CREDENTIALS_UNSAFE] (MEDIUM): The skill encourages users to store sensitive plain-text credentials (ZOOM_PASSWORD, SERVICE_PASSWORD) in environment variables. While the skill claims these are only stored locally, they become accessible to any process or AI agent running in that environment.
[COMMAND_EXECUTION] (HIGH): The 'Error Handling' and 'Authentication Flow' sections describe a process where the agent handles 2FA by reading codes from 'iMessage notification'. Accessing a user's private messages typically requires excessive system permissions and poses a severe privacy risk.
[PROMPT_INJECTION] (LOW): The 'Self-Improvement Instructions' section directs the agent to 'Update skill with new webinar features' and 'Log successful scheduling patterns' based on its own observations, which is a form of self-modifying behavior that could be exploited via indirect injection if it encounters malicious UI elements.

zoom