openrouter
Pass
Audited by Gen Agent Trust Hub on Oct 20, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill documents and enforces security best practices, including redacting credentials from logs, using server-side secret management for API keys, and avoiding the collection of private user data in telemetry.
- [SAFE]: External references and API endpoints utilize official domains for OpenRouter and OpenAI, which are recognized as well-known and trusted technology services.
- [INDIRECT_PROMPT_INJECTION]: The skill provides instructions for processing data from external API responses, which are considered untrusted inputs.
- Ingestion points: Untrusted data enters the context via Chat Completion responses and tool call results as described in references/reliability.md.
- Boundary markers: The instructions focus on structural validation but do not specify LLM-level delimiters for output processing.
- Capability inventory: No high-risk capabilities, such as arbitrary command execution or unrestricted file access, are requested or configured within the skill's metadata.
- Sanitization: The guidelines recommend strict schema parsing and task-specific validation before accepting and processing API results, as noted in references/reasoning-and-output.md.
Audit Metadata