The Agent Skills Directory

Indirect Prompt Injection Surface: The skill is designed to process external data, including campaign briefs and product catalogs from connected storefronts.
Ingestion Points: The skill reads external data in SKILL.md (Pre-flight and Stage 1 sections) where it fetches content briefs and product lists from storefronts like Shopify or Square.
Potential Concern: Data originating from these external sources could potentially contain embedded instructions aimed at influencing the AI agent's behavior during the content generation process.
Capability Inventory: Across its scripts and instructions, the agent has the ability to read local files, perform network operations via WebFetch (to Canva and HubSpot), and utilize specialized tools for design management.
Sanitization and Mitigation: The skill utilizes a robust "Approval Gate" architecture. It explicitly pauses for owner confirmation at multiple checkpoints (Checkpoints 1, 2, and 3) before proceeding to subsequent stages or executing final API calls. This ensures that any generated content or planned actions are reviewed by a human before execution.
Communication with Well-Known Services: The skill performs network operations to interact with Canva and HubSpot.
Description: It uses the WebFetch tool to communicate with api.canva.com and api.hubapi.com for asset management, design generation, and social media scheduling.
Context: These are established marketing platforms. The skill follows standard authentication practices (OAuth 2.0 Bearer tokens) and includes mechanisms like rate-limiting, budget estimation, and visual verification of exports to ensure data integrity and prevent service disruption.

canva-creator