The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute a Python script at the path .agents/scripts/auto_preview.py for previewing deployments. This script is not included in the provided 20 files, making its contents and safety profile unverifiable.
[DYNAMIC_EXECUTION]: The orchestrator facilitates the generation and runtime execution of code across various tech stacks, which is a key component of its functionality but poses a risk if generated code contains malicious logic.
[INDIRECT_PROMPT_INJECTION]: The skill translates natural language requests into complex project plans and executable code, which represents a significant surface for indirect prompt injection. Malicious instructions embedded in a user's project description could influence sub-agents to generate insecure code or perform unauthorized file/network operations.
Ingestion points: User's natural language requests processed by the app-builder and project-detection components.
Boundary markers: Not explicitly specified in the coordination instructions for agent-to-agent communication.
Capability inventory: The orchestrator and specialist agents have broad permissions including Read, Write, Edit, Bash, and Agent tools.
Sanitization: The files do not describe specific sanitization or validation routines for untrusted input before it is used for code generation.

app-builder