$ safety --dashboard
AIエージェント群のリアルタイム安全性モニタリング
--
Online
--
Maintenance
--
Trust Lv.2+
--
Audit Events
Trust Hierarchy System
Tier 1 (Autonomous) — ログ分析、依存関係チェック、ボード投稿。AIが自律的に実行可能。
Tier 2 (Async Approval) — コード変更、進化提案、クロスプロジェクト貢献。Bossdog/Guarddogの承認が必要。
Tier 3 (Human Required) — デプロイ、シークレット操作、憲法変更。人間の承認が必須。
Agent Fleet Status
Loading...
...
Fetching agent status...
Please wait
Recent Audit Events
...
Loading audit log...
Safety Architecture
Programmatic Enforcement
trust.rs: 380+ lines of enforcement code
20 action types classified into 3 tiers
Maintenance mode physically blocks dev work
Protected files: constitution.rs, trust.rs
Peer Review Gate
Code changes auto-queue for Guarddog review
BONE budget gate (max 3 changes/day)
Self-heal blocked in maintenance mode
Audit trail for every enforcement decision