The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill involves the dynamic generation and publication of React/HTML code. The workflow uses canvas-source-retrieve to read and canvas-publish-create to deploy code to a sandboxed environment.\n- [COMMAND_EXECUTION]: Instructions include using curl from a shell to upload media to a platform-provided upload_url. This facilitates local file interactions and network requests directed at the vendor's infrastructure.\n- [INDIRECT_PROMPT_INJECTION]: The skill builds apps that process data, creating a vulnerability surface.\n
Ingestion points: Data enters via ph.query, ph.loadInsight, and ph.connectors.call in SKILL.md and the generated src/canvas.tsx.\n
Boundary markers: The instructions do not specify the use of delimiters or ignore markers for untrusted data.\n
Capability inventory: The system supports canvas-edit-create for code modification in SKILL.md and ph.tools.call for executing platform tools.\n
Sanitization: No specific sanitization or escaping mechanisms for external content are mandated before processing.

building-canvases