The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the creation of automation scenarios that ingest data from untrusted external sources, such as Gmail attachments and payment gateway webhooks. This data is then processed by AI Agents for tasks like document classification and customer routing. This architecture creates a surface for indirect prompt injection, where an attacker could embed malicious instructions within an invoice or support request to manipulate the agent's logic.
Ingestion points: Gmail "Watch Emails" trigger (SKILL.md), Cardcom/Tranzila/Grow webhooks (SKILL.md), and HTTP module responses from external APIs.
Boundary markers: The instructions do not explicitly mandate the use of robust delimiters or "ignore" instructions for the AI Agent prompts when interpolating external data.
Capability inventory: The skill utilizes capabilities for network operations (WhatsApp, SMS), file management (Google Sheets, Morning), and document generation across various scripts and references.
Sanitization: While basic string operations like toString and trim are mentioned for data handling, they are not presented as security measures to prevent prompt injection.
[EXTERNAL_DOWNLOADS]: The instructions rely on several community-built Make.com modules, such as "Morning by Callbox" and the "Hebcal" module. These are external dependencies hosted on the Make.com marketplace. The skill also directs the agent to interact with numerous third-party Israeli service APIs for accounting, ERP, and messaging.
Source URLs: References include apps.make.com/hebcal-ryuwr8, apps.make.com/green-invoice-y563os, and various Israeli service domains (e.g., api.greeninvoice.co.il, api.inforu.co.il).
Context: These references are central to the skill's purpose of enabling Israeli business integrations and are documented as third-party modules.

make-com-israeli-automations