policy
permission_mode, sandbox_policy, and permission_profile.
Counts of explicit source-field observations. Values stay inside a closed vocabulary, with other when the source is outside it. Command text is never stored. A grant is not effective authority. An empty lane is not a measured zero.
All-time counts — window_from/window_to label the snapshot but do not filter.
Closed Security Vocabulary Lanes: Evaluates security stance across 6 deterministic dimensions:
Configured permission_mode, sandbox_policy, and permission_profile governing agent boundaries.
Recorded approval decisions and prefix permissions. A grant does NOT infer effective authority without explicit verification.
Observed dangerous command categories (e.g. force_push, privilege escalation) without persisting raw command strings.
Command line arguments, environment variables, and authentication tokens are strictly omitted from storage.
Closed vocabulary standards and compliance boundaries:
Total recorded security telemetry events in observation window
Peak session concurrency sharing identical permission settings
Not inferred; requires policy proof to transition from 'Not assessed'
Raw commands are strictly omitted from all persistence layers
permission_mode, sandbox_policy, and permission_profile.
permission_decision and approved_prefix_count. A grant does not confer effective authority.
approval_policy. This is an observed field, not a review verdict.
dangerous_command category and sandbox_override_requested. The command line is omitted.
Not inferred. Without a policy layer the assessment stays not assessed.
Not stored. Absence of an effect row does not mean the action had no effect.
| Lane | Harness | Kind | Value | Observations | Sessions |
|---|
SELECT * FROM v_security_posture ORDER BY observation_count DESC, harness, kind;
Recorded window: —
Total rows: —