subagents-management

Pass

Audited by Gen Agent Trust Hub on May 9, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses Python scripts to perform file system operations such as listing, creating, moving, and deleting markdown and TOML configuration files in user and project directories (~/.claude/agents/, ~/.codex/agents/, etc.).
  • [COMMAND_EXECUTION]: Includes a deletion script (delete_subagent.py) with a --force flag. However, the skill instructions explicitly mandate using the AskUserQuestion tool for manual user confirmation before any deletion occurs, mitigating the risk of accidental or unauthorized file removal.
  • [PROMPT_INJECTION]: The skill presents an indirect prompt injection surface as it allows the agent to generate subagent configuration files containing system prompts derived from user-supplied input. If a user provides a malicious prompt, it will be saved as the subagent's instruction set, which could influence the behavior of that subagent when invoked later.
  • Ingestion points: User-provided text via the --prompt and --description arguments in create_subagent.py.
  • Boundary markers: No explicit boundary markers or protection against embedded instructions are used in the generated subagent markdown files.
  • Capability inventory: Scripts for file creation, movement, and deletion are present (create_subagent.py, move_subagent.py, delete_subagent.py).
  • Sanitization: Basic alphanumeric validation is performed on the subagent name, but the prompt content is interpolated without sanitization into the final configuration file.
Audit Metadata
Risk Level
SAFE
Analyzed
May 9, 2026, 10:57 PM