The Agent Skills Directory

[DATA_EXFILTRATION]: The script scripts/send_email.py allows reading arbitrary files from the local filesystem using the --attachment and --html arguments (lines 79 and 96). Since the skill does not implement path validation or a restricted directory whitelist, an agent could be instructed to read and email sensitive files such as SSH keys, configuration files, or environment variables to an external SMTP server.
[INDIRECT_PROMPT_INJECTION]: The skill is vulnerable to indirect injection as it ingests data for the email body and recipients without sanitization. Mandatory evidence chain: 1) Ingestion points: The --to, --subject, --body, and --html arguments in scripts/send_email.py. 2) Boundary markers: None provided to distinguish untrusted data from instructions. 3) Capability inventory: Network transmission via smtplib and arbitrary file reads via the Path.open method in scripts/send_email.py. 4) Sanitization: No filtering or validation of file paths or email content is performed.
[EXTERNAL_DOWNLOADS]: The skill's metadata in SKILL.md references the official Nodemailer repository on GitHub, which is a well-known and established open-source project for email automation.

email-sender