The Agent Skills Directory

[PROMPT_INJECTION]: The skill functions by dynamically loading instructions from markdown files and passing them to subagents. This architecture is the core mechanism of the tool, allowing it to aggregate specialized behaviors.
Ingestion points: The skill reads .md files from the ./agents/ and ~/.claude/agents/ directories as defined in the configuration section of SKILL.md.
Boundary markers: The skill constructs prompts using the template "{agent file content}\n\nTask: {task description}" without additional isolation markers.
Capability inventory: The skill uses an internal subagent tool to execute the instructions found in the scanned files.
Sanitization: Content from the markdown files is treated as trusted instructional data for the subagents, which aligns with the tool's primary purpose of team orchestration.

team-builder