role-creator

Warn

Audited by Gen Agent Trust Hub on Mar 22, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill generates and executes a shell command (agent-team role create) by interpolating user-provided text into command flags such as --description and --system-goal. Without explicit sanitization or escaping of shell metacharacters, this pattern presents a command injection vulnerability.
  • [DATA_EXFILTRATION]: The skill attempts to access the ~/.claude/skills/ directory to check for existing roles. Reading from hidden application configuration directories in the user's home path is a sensitive operation that can expose details of the local environment.
  • [PROMPT_INJECTION]: This skill provides a surface for indirect prompt injection by generating system instructions and skill metadata from unvalidated user input.
  • Ingestion points: User-provided natural language fields for role description and system goals (Step 2).
  • Boundary markers: The template structure for the generated system.md and SKILL.md files lacks explicit delimiters or instructions to ignore embedded commands.
  • Capability inventory: The skill can execute the agent-team CLI and write files to the repository.
  • Sanitization: Input validation is limited to normalizing the role name to kebab-case, with no sanitization for descriptions or goals.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Mar 22, 2026, 06:24 PM