Safe execution boundaries before autonomy.
The prototype favors explicit permissions, blocking gates and inspectable evidence over unrestricted agent execution.
Current security posture
Skillaryx V0.1 is a local prototype. The security model is conservative because agent capabilities can touch files, shell commands, tools and external services.
Deny by default
Network access is denied by default. Shell execution is constrained to explicit allowlists. Model responses are treated as data and are not blindly executed as code.
Execution gates
Blocking evaluations and policy checks must pass before runtime execution is permitted. The failure-path prototype verifies that a failed blocking eval prevents runtime invocation.
Secrets
Provider credentials are supplied through environment configuration, excluded from source control and should never be returned to the frontend or intentionally logged.
Responsible disclosure
Report potential security issues privately to founder@skillaryx.com. Please avoid publishing sensitive exploit details before there has been a reasonable opportunity to investigate.
Scope note
This page describes the current local prototype and public website. It is not a claim of penetration testing, SOC 2, ISO 27001 or other formal certification.