Security audit
astrai-inference-router
Security checks for vulnerabilities and agentic risk
Overview
The plugin generally does what it says (routes LLM calls through Astrai and requires an Astrai API key), but the SKILL.md makes strong privacy claims (local PII stripping, zero retention, EU-only enforcement) that are not implemented in the included code — this mismatch warrants caution before installing.
This skill will route your agent's LLM requests through a third-party gateway (as-trai.com) and requires giving that service an API key. The documentation promises local PII stripping, EU-only enforcement, and zero retention in some modes, but the included plugin does not perform any prompt sanitization — it merely sets headers and sends requests to the remote gateway. Before installing, verify the upstream service and repository: 1) review the GitHub source and confirm the remote domain and project legitimacy; 2) if you require true local PII stripping, do not rely on this skill unless you add/verify client-side sanitization code; 3) consider testing in a sandboxed environment and monitor outgoing traffic to confirm behavior; and 4) be aware that providing your ASTRAI_API_KEY gives the gateway the ability to process your prompts and bill usage. If the privacy guarantees are critical, request proof or code changes that implement local stripping and enforce region constraints locally rather than relying on headers sent to the remote service.
SkillSpector findings are pending for this release.