The Agent Skills Directory

REMOTE_CODE_EXECUTION (CRITICAL): The skill requires users to run unverified scripts piped to bash from an untrusted domain (canifi.com) for both skill installation and environment setup. Evidence: curl -sSL https://canifi.com/skills/outlook/install.sh | bash and curl -sSL https://canifi.com/install.sh | bash.
EXTERNAL_DOWNLOADS (HIGH): The installation process fetches code from external, non-trusted sources without any integrity checks or version pinning.
CREDENTIALS_UNSAFE (HIGH): The skill explicitly advises users to store their email passwords in the SERVICE_PASSWORD environment variable. This plaintext credential storage is accessible to the agent and other local processes.
PROMPT_INJECTION (HIGH): The skill is highly vulnerable to Indirect Prompt Injection (Category 8). 1. Ingestion points: External email content from Microsoft Outlook web interface. 2. Boundary markers: Absent; there are no instructions to the agent to distinguish between user commands and data-embedded instructions. 3. Capability inventory: Extensive write privileges including composing, sending, replying, and folder management. 4. Sanitization: Absent; the agent processes raw email content directly without filtering.
DATA_EXFILTRATION (HIGH): The combination of unverified remote code execution and full access to private emails (including send/forward capabilities) provides multiple high-risk paths for automated data theft.

outlook