dynamic-router
Pass
Audited by Gen Agent Trust Hub on Aug 17, 2026
Risk Level: SAFEPROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user tasks and routes them to sub-agents based on complexity, which creates a surface where instructions embedded in the data could potentially influence the agent's logic.
- Ingestion points: User task strings supplied through the
/dynamic-taskcommand or theconductprogrammatic function defined inSKILL.md. - Boundary markers: No explicit delimiters or safety instructions (e.g., "ignore embedded instructions") are present in the provided snippets to isolate user tasks.
- Capability inventory: The skill identifies and routes tasks to adapters capable of performing file lookups, code implementation, and system architecture planning.
- Sanitization: No input validation or sanitization logic is shown in the analyzed documentation or code snippets.
- [DATA_EXPOSURE_AND_EXFILTRATION]: The skill indicates use of external AI provider endpoints to facilitate task execution across its various model adapters.
- Evidence: The skill documentation references connectivity to the Kimi API (
api.moonshot.cn/v1) for model access within the OpenCode and Pi adapters.
Audit Metadata