The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted agent definition files and user task descriptions, which are then interpolated into sub-agent prompts without clear boundaries or sanitization. Ingestion points: Reads agent markdown files from the local project directory (./agents/) and the global user agent directory (~/.claude/agents/), as well as user-provided task descriptions. Boundary markers: Absent. Instructions are concatenated directly: {agent file content}\n\nTask: {task description}. Capability inventory: Spawns parallel sub-agents using the Agent tool with subagent_type: 'general-purpose', providing a broad range of capabilities to potentially hijacked sub-agents. Sanitization: None. The skill does not escape or validate the content of the markdown files or the task description before passing them to the sub-agent tool.
[COMMAND_EXECUTION]: The skill instructs the agent to run the claude agents CLI command to discover available agents. While this is a platform-specific discovery command, it involves executing shell commands based on instructions in the skill.

team-builder