The Agent Skills Directory

[COMMAND_EXECUTION]: The skill orchestrates project setup by issuing shell commands including 'mkdir', 'cd', and 'git' based on user-provided project names and paths. While these are necessary for project initialization, they represent a direct interaction with the host filesystem that relies on the agent's ability to sanitize paths correctly.
[PROMPT_INJECTION]: The skill processes untrusted user input (startup ideas and requirements) and passes this data through a multi-stage pipeline involving other specialized agents and skills. This creates an indirect prompt injection surface where malicious instructions embedded in a project description could potentially influence the behavior of downstream code-generation tools. Ingestion points: Project vision and requirements gathering phases (Phases 1 and 2). Boundary markers: Lacking explicit delimiters or safety instructions in the prompts that pass user data to sub-skills like 'senior-architect' or 'code-assistant'. Capability inventory: The workflow can create filesystem structures and generate source code. Sanitization: No content filtering or natural language sanitization is performed on the user-provided inputs before they are interpolated into prompts for subsequent processing steps.

studio-startup