The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The 'automation' artifact template provides webhook triggers for Slack, Telegram, and other connectors that ingest external user data. In the provided 'workflow.ts' file, this untrusted data is interpolated directly into an AI agent's prompt without the use of boundary markers or input sanitization. This creates an architecture vulnerable to indirect prompt injection, where malicious instructions in external messages could manipulate the agent's behavior.
Ingestion points: Webhook handlers in 'artifacts/automation/files/src/triggers/slackTriggers.ts', 'telegramTriggers.ts', and 'exampleConnectorTrigger.ts'.
Boundary markers: Absent in the prompt construction within 'artifacts/automation/files/src/mastra/workflows/workflow.ts'.
Capability inventory: Subagent generation capabilities, network access via integrated service triggers, and background task orchestration via Inngest.
Sanitization: No escaping or filtering of external content is implemented in the template logic.
[DYNAMIC_EXECUTION]: The 'mockup-sandbox' scaffold includes a Vite plugin ('mockupPreviewPlugin.ts') that performs runtime discovery of component files and generates a module mapping paths to dynamic 'import()' calls. While this is a standard implementation for a development-time component gallery, it involves dynamic loading from computed paths based on file system state.
[EXTERNAL_DOWNLOADS]: The 'expo' artifact's landing page template ('landing-page.html') fetches the 'qr-code-styling' library from the well-known 'unpkg.com' CDN to generate QR codes for mobile previewing of the scaffolded application.

artifacts