The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to handle external data that may be controlled by third parties, representing a potential surface for indirect prompt injection.
Ingestion points: The skill processes data from Canva API endpoints, including design comments (/comment_threads) and webhook payloads, which can contain arbitrary user-generated content.
Boundary markers: While the skill focuses on data processing logic, it does not explicitly define delimiters for separating untrusted content when passed back to the agent.
Capability inventory: The skill is configured with Read, Write, and Edit tool access and demonstrates database interactions for data cleanup and retention.
Sanitization: The skill implements strong defensive measures, including a redactCanvaData utility to prevent sensitive tokens from entering logs and instructions to scan uploaded assets for malware.
[SAFE]: The skill provides best-practice implementation for secure data handling.
The redactCanvaData function effectively prevents common data leakage patterns by stripping access tokens and secrets before logging.
The CanvaUrlTracker ensures that time-limited signed URLs are not cached beyond their validity period, preventing unauthorized access to stale resources.
The data retention and deletion workflows align with security principles for minimizing data exposure and meeting compliance requirements.

canva-data-handling