Autonomous Workflow Exfiltration Risk & Sandbox Containment Index
Verified ModelComputes the composite security exposure score (0 to 100) for GitHub Actions and local MCP agent loops based on container isolation multiplier, maximum permitted tool invocations, untrusted third-party skills, and subprocess timeout window.
- R_score (Composite Risk Index)
- 11.2 / 100
- W_sandbox (Isolation Multiplier)
- 0.30x
- N_calls (Max Tool Invocations)
- 18
- E_code (Predicted Exit Status)
- Exit 0 (Verified)
Deterministic exposure rating; values >= 65 trigger mandatory Exit 78 lockdown.
0.30 for strict rootless chroot+seccomp, 0.65 for Docker bridge, 1.45 for raw host.
Upper bound on consecutive MCP tool calls before human-in-the-loop gate.
Resolved harness termination signal under current policy vector.