The Agent Skills Directory

[DATA_EXPOSURE]: The skill implements a 'Secret tripwire' using regex to scan the subject and body of every email for sensitive patterns (e.g., AWS access keys, Resend API keys, GitHub tokens, and private keys). If any are detected, the send operation is blocked, effectively preventing accidental credential exfiltration.- [EXTERNAL_DOWNLOADS]: The skill performs authorized network requests to 'api.resend.com', which is a well-known transactional email service. This communication is the primary purpose of the skill and is handled via a secure platform tool.- [COMMAND_EXECUTION]: Shell commands and a Python script are used for internal logic, such as building the JSON payload and validating recipient formats. These scripts are static, self-contained, and do not execute external or user-provided code.- [PROMPT_INJECTION]: The skill includes explicit instructions for the agent to treat all fetched context as untrusted and specifically warns against allowing external data to inject instructions into the email body.- [SAFE]: The skill employs multiple 'fail-closed' mechanisms, including a kill-switch, daily budget caps, and a shared ledger for deduplication, ensuring that if any safety check fails, the email is not sent.

send-email