agentgate-proxy v0.1.0 trace inspector
SELECT TEST VECTOR 6 calibrated cases
INBOUND ASSESSMENT REQUEST POST /v1/assess
GATEWAY EVALUATION TRACE
28ms (p95 <38ms)
HTTP VERDICT
403 FORBIDDEN · DENY
Action blocked: untrusted prompt influence exceeds safety threshold (0.50).
CALIBRATED RISK HEADS (SINGLE FORWARD PASS)
HEAD CLASS PROB STATUS
untrusted_influence likely 0.94 [TRIGGER]
goal_alignment misaligned 0.91 [FLAGGED]
reversibility irreversible 0.85 [HIGH]
external_send no 0.96 [OK]
MATCHED POLICY RULE (policy.yaml)
tools.bash.rules[0]:
  when: { head: untrusted_influence, option: likely, gt: 0.50 }
  action: deny
  reason: untrusted_prompt_influence