The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from user-supplied arguments and external context about recipients to compose email bodies.
Ingestion points: The ${var} input and recipient-specific context from external sources.
Boundary markers: The skill does not use explicit delimiters or markers when interpolating untrusted data into the composition prompt, though it does instruct the agent to treat fetched context as untrusted.
Capability inventory: The skill possesses network capabilities (via ./secretcurl) and filesystem write access (to logs and drafts).
Sanitization: A regex-based "Secret tripwire" is implemented to identify and block the transmission of sensitive tokens (e.g., AWS, GitHub, Resend) or private keys if they appear in the email body.
[COMMAND_EXECUTION]: The skill executes various local CLI tools including python3, jq, grep, and date, as well as a local script scripts/check_email_bounces.py. These executions are limited to generating JSON payloads, maintaining the email ledger, and checking delivery statuses.
[EXTERNAL_DOWNLOADS]: The skill communicates with the Resend API (api.resend.com) to send emails and poll for bounces. Resend is a well-known service, and these interactions are essential for the skill's primary function.
[SAFE]: The skill follows security best practices for handling secrets, such as avoiding passing the RESEND_API_KEY as a command-line argument by using environment variables inside Python heredocs and utilizing placeholders for shell-level secret expansion.

send-email