The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill instructs the agent to collect sensitive credentials (SMTP passwords, authorization codes, and API keys) and passes them as plain-text command-line arguments to the send_email.py script. This exposes secrets to process lists and system logs. Additionally, the skill uses a hardcoded default email address (agent_skill_test@126.com) and explicitly directs the agent to use it without asking the user, then solicits the SMTP authorization code for this specific account.
[DATA_EXFILTRATION]: The skill provides the capability to attach arbitrary files from the local filesystem using the --attach parameter. This can be abused to exfiltrate sensitive data by emailing local files to an external recipient.
[INDIRECT_PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it processes untrusted user content and template variables which are then transmitted via network operations (SMTP).
Ingestion points: User-provided email body (--content) and template variables (--template-vars) in send_email.py.
Boundary markers: The instructions do not define clear delimiters or safety warnings for the agent when processing user-supplied content.
Capability inventory: The send_email.py script has the ability to read arbitrary files from the disk and send data over the network (SMTP).
Sanitization: The script performs no validation or sanitization of the email content or file paths before processing them.

send-email