add-openrouter
Pass
Audited by Gen Agent Trust Hub on Sep 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill fetches a comprehensive list of models and their metadata from an external API (openrouter.ai) to populate a local catalog stored in SQLite. If the remote API provides malicious instructions within model names, descriptions, or parameters, these could influence the agent's behavior during model selection or task execution.
- Ingestion points: Fetches model data from
https://openrouter.ai/api/v1/modelsin Step 3 and Step 5 (SKILL.md). - Boundary markers: The instructions do not define delimiters or specific "ignore" commands for the ingested model metadata.
- Capability inventory: The skill utilizes
write_file,read_file, andrun_codetools, providing a surface for local file modification and code execution based on processed data (SKILL.md). - Sanitization: There is no mention of sanitizing or validating the contents of the model catalog before it is stored in SQLite or used to update configuration files.
Audit Metadata