The Agent Skills Directory

[DATA_EXFILTRATION]: The send_email.py script implements functionality to attach local files to outgoing emails via the --attachments argument. This creates a surface for data exfiltration if an agent is directed to attach sensitive files, such as SSH keys, environment files, or application secrets, and send them to an external email address.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes untrusted user instructions (e.g., which files to attach and where to send them) without specific guardrails. An attacker could use malicious input to trick the agent into performing unauthorized data transfers using the skill's legitimate email-sending capabilities.
Ingestion points: User-provided arguments for recipients, subject, body, and file attachment paths in SKILL.md.
Boundary markers: None identified in the prompt templates or instructions to prevent the agent from processing malicious file paths.
Capability inventory: File system read access (via open() in scripts/send_email.py) and network transmission (via smtplib in scripts/send_email.py).
Sanitization: The script validates file existence but does not restrict access to sensitive directories or perform path sanitization to prevent access to system-critical files.

smtp-send