The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest inbound email content, which constitutes untrusted external data capable of carrying malicious instructions.
Ingestion points: Webhook endpoints in SKILL.md and references/security-levels.md receive email.received events containing potentially malicious payloads.
Boundary markers: The skill explicitly documentation and implements "Strict Allowlists", "Domain Allowlists", and recommends stripping quoted threads to isolate new content.
Capability inventory: The skill utilizes the Resend API for reading and sending emails, with specific implementation levels (Level 4/5) aimed at restricting these capabilities for untrusted senders.
Sanitization: Includes code for stripping HTML, truncating content to prevent token stuffing, and filtering based on safety patterns.
[EXTERNAL_DOWNLOADS]: Recommends the installation of Tailscale for secure local tunneling using an installation script from the vendor's official domain.
Evidence: curl -fsSL https://tailscale.com/install.sh | sh in references/webhook-setup.md.
[PRIVILEGE_ESCALATION]: Instructions include the use of sudo for administrative tasks related to tunnel configuration.
Evidence: Commands such as sudo tailscale up and sudo tailscale funnel 3000 are suggested in references/webhook-setup.md.

agent-email-inbox