The Agent Skills Directory

[Indirect Prompt Injection Surface]: The skill processes user-provided requirements through the $ARGUMENTS variable and reads structured plan files like canvas-app-plan.md to direct the creation of application screens. Because this data influences the agent's logic and code generation without explicit sanitization or boundary markers, it represents a surface where instructions could potentially be introduced.\n
Ingestion points: $ARGUMENTS (SKILL.md), canvas-app-plan.md (SKILL.md), canvas-app-shared.md (SKILL.md).\n
Boundary markers: None identified in the instruction text.\n
Capability inventory: Write, Edit, and Bash for directory operations (SKILL.md).\n
Sanitization: No explicit sanitization or validation logic is defined for the external inputs.\n- [Command Execution]: The skill uses the Bash tool to perform workspace initialization, specifically creating directories based on names derived from user requirements. While used for routine file system management, executing shell commands influenced by external input is a security consideration that warrants oversight to prevent injection.\n
Evidence: The skill specifies using Bash to create folders during the 'Establish the Workspace' step in SKILL.md.

canvas-app