The Agent Skills Directory

[DATA_EXFILTRATION]: The skill reads from the ~/.claude/teams/ directory to retrieve agent configurations and prompts. Accessing these system-level directories can expose sensitive operational data and agent instructions.
[COMMAND_EXECUTION]: The skill utilizes the Bash tool to execute mkdir -p and mv commands. These commands incorporate variables like {save_name} which are derived from user input or arguments. Without proper validation, these inputs could be manipulated to execute unintended shell commands.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection and path traversal because it uses unsanitized user arguments to construct file paths for both reading configurations and writing snapshots.
Ingestion points: Arguments passed to the skill ($ARGUMENTS) and user input from the AskUserQuestion tool.
Boundary markers: There are no markers or validation steps to ensure that provided names do not escape the intended directory structure.
Capability inventory: File system access via Bash (mkdir, mv), Write, and Edit tools.
Sanitization: The skill does not perform any escaping or validation on user-provided strings before using them in file paths or commands.

team-save