The Agent Skills Directory

[COMMAND_EXECUTION]: The skill orchestrates sub-agents by retrieving persona instructions from external markdown files and executing them via the platform's Agent tool.
[DATA_EXPOSURE]: Scans for agent definitions in the project-local ./agents/ folder and the global ~/.claude/agents/ configuration directory.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it interpolates the raw, unsanitized content of markdown files directly into the instructions for spawned sub-agents.
Ingestion points: Reads markdown files from ./agents/**/*.md and ~/.claude/agents/**/*.md.
Boundary markers: None; content is interpolated using a simple template: "{agent file content}\n\nTask: {task description}".
Capability inventory: Uses the Agent tool to spawn sub-agents which may have code execution or file system access capabilities.
Sanitization: No validation or sanitization is performed on the content of the markdown files before execution.

team-builder