The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's architecture relies on synthesizing content produced by multiple parallel runner agents, introducing a surface for indirect instructions to influence the final design.
Ingestion points: The skill collects design packages from multiple child agents via task_status in Phase B of SKILL.md.
Boundary markers: While child agents are isolated from the parent's context, there are no specific delimiters or sanitization steps mentioned for the synthesized output.
Capability inventory: The skill possesses the capability to delegate tasks to other agents, write to the filesystem, and execute project-specific setup commands.
Sanitization: No explicit filtering or validation of the text/code generated by the child runners is performed before it is integrated into the workspace.
[COMMAND_EXECUTION]: The skill executes the kickoff-setup command if the necessary configuration files (kickoff.yaml) are absent, which constitutes an automated execution of a local tool.
[DYNAMIC_EXECUTION]: The skill uses delegate_task to spawn and manage child agents dynamically, resolving roles and models at runtime based on the workspace configuration and orchestrator capabilities.

architect