model-routing-patterns
Pass
Audited by Gen Agent Trust Hub on Sep 17, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill consists of configuration templates and local validation scripts intended for optimizing OpenRouter API usage. The included bash scripts (e.g., validate-routing-config.sh, analyze-cost-savings.sh) process JSON configuration files locally using standard utilities like jq and bc.\n- [INDIRECT_PROMPT_INJECTION]: The skill defines patterns for routing requests based on prompt complexity, which is a target for indirect prompt injection where crafted input might influence model selection costs. This is an intended architectural feature and the skill provides documentation for monitoring and budget alerts to manage this risk.\n
- Ingestion points: RequestContext object in templates/routing-config.py and templates/routing-config.ts.\n
- Boundary markers: No explicit prompt delimiters are implemented in the routing classification logic.\n
- Capability inventory: The logic influences model selection and subsequent network calls to the OpenRouter API.\n
- Sanitization: No sanitization is performed on the prompt content used for complexity classification.\n- [DYNAMIC_EXECUTION]: The balanced-routing.json template references a .pkl file for machine learning-based classification. While Pickle files represent a potential unsafe deserialization vector, the skill does not provide any code that actually implements the loading or execution of this file, serving only as a template placeholder.
Audit Metadata