The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes the claude agents CLI command at runtime to discover a list of available agents. This is a platform-native command used for inventory, but it involves direct shell invocation.
[INDIRECT_PROMPT_INJECTION]: The skill reads external markdown files from local directories and interpolates their contents directly into subagent prompts without sanitization.
Ingestion points: Markdown files located in ./agents/**/*.md and ~/.claude/agents/**/*.md.
Boundary markers: None. The file content is interpolated as {agent file content} directly into the prompt provided to the Agent tool.
Capability inventory: The skill utilizes the Agent tool to spawn subagents, which may have their own sets of capabilities (file access, code execution).
Sanitization: No validation, escaping, or sanitization is performed on the content read from the agent files before it is passed to the LLM.
[DYNAMIC_EXECUTION]: The skill dynamically generates instructions for the Agent tool by concatenating file content with user task descriptions at runtime (prompt: "{agent file content}\n\nTask: {task description}").

team-builder