The Agent Skills Directory

[DATA_EXFILTRATION]: The script scripts/smtp_send.py implements a workflow that reads local files from the filesystem (via the --attach argument in the read_attachments function) and transmits them to external servers using SMTP (send_via_smtp) and IMAP (append_to_sent_mailbox). This combination of local file access and network transmission constitutes a capability for data exfiltration if misused.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits a vulnerability surface for indirect prompt injection attacks because it ingests external data that influences its actions.
Ingestion points: Data enters the skill through the --to, --subject, --body, and --attach command-line arguments in scripts/smtp_send.py.
Boundary markers: There are no explicit boundary markers or instructions to the agent to treat the email content or attachment paths as untrusted data.
Capability inventory: The skill possesses file-read capabilities (path.read_bytes()) and network-send capabilities (SMTP and IMAP protocols) within scripts/smtp_send.py.
Sanitization: The skill does not perform path validation or content filtering to prevent access to sensitive directories (e.g., .ssh or .aws).
[EXTERNAL_DOWNLOADS]: The documentation in SKILL.md and references/env.md instructs the user to install the imapclient Python package from a public registry if they wish to use the IMAP sync feature.
[COMMAND_EXECUTION]: The workflow involves the agent executing the scripts/smtp_send.py script with various flags, which allows the agent to interact with the host's network and filesystem.

email-smtp-send