The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface as it processes agent-generated HTML prototypes and user-supplied feedback. Ingestion points: Prototype HTML files, board.json manifest, and feedback.json. Boundary markers: Prototypes are rendered within a sandboxed iframe and the system injects a strict Content Security Policy (default-src 'none'; connect-src 'none'; frame-src 'none') to prevent data exfiltration. Capability inventory: The skill uses a Python script to write files and provides shell instructions for serving the board. Sanitization: The build_board.py script validates HTML structure, blocks external resource links, and employs a custom CSS decoder to detect and reject @import rules, even when obfuscated with escapes.
[COMMAND_EXECUTION]: The skill guides the agent to execute a local Python build script and a project-specific serve utility. The build script implements path traversal prevention using Path.relative_to to ensure variants are resolved within the intended directory.
[SAFE]: No malicious behavior, obfuscation for evasion, or unauthorized network operations were identified. All functions are transparently documented and support the skill's primary purpose.

interaction-design-board