create-agent
Pass
Audited by Gen Agent Trust Hub on Sep 27, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied agent names, categories, and responsibilities to generate new sub-agent instruction files (.md).\n
- Ingestion points: User arguments and interactive responses captured in Step 1 of SKILL.md.\n
- Boundary markers: The delegation prompt in Step 2 uses bullet points and labels (e.g., 入力:, 出力先:) to separate instructions from data.\n
- Capability inventory: The skill (or its delegated sub-agent) utilizes Write, Edit, and Bash to create and move files.\n
- Sanitization: The skill relies on model-based validation in Steps 3 and 4 to verify the generated frontmatter and permissions, but lacks strict structural sanitization of the user input itself.\n- [COMMAND_EXECUTION]: The skill performs shell-level operations, specifically moving files (mv) from temporary staging areas (_/dotclaude/) to project configuration directories (.claude/agents/) as part of the dotclaude-via-temp workflow.
Audit Metadata