The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill loads content from external markdown files (agent personas) and user-provided task descriptions to construct a prompt for subagents. The absence of boundary markers or sanitization could allow malicious instructions within an agent file to hijack the subagent's execution.
Ingestion points: Reads files from ./agents/**/*.md and ~/.claude/agents/**/*.md as specified in the 'Configuration' section.
Boundary markers: None; the instructions use a simple concatenation: {agent file content}\n\nTask: {task description} in Step 4.
Capability inventory: Utilizes the Agent tool to spawn general-purpose subagents in parallel.
Sanitization: No validation or filtering is performed on the loaded content or user input before it is interpolated into the subagent prompt.
[COMMAND_EXECUTION]: The skill instructs the agent to execute the claude agents command to discover agents available via the CLI, as described in Step 1.

team-builder