Agent security framework for pi. Immutable audit trail, permission policies, destructive command guard, self-modification detection.

Based on 0DIN research: "Context is the control plane." Detects self-modification attacks (Palisade Research).

npm install -g @artale/pi-sentinel

• 22 destructive command patterns blocked via tool_call hook
• Immutable audit trail with SHA-256 hashed entries
• Permission policies — allow/deny rules for paths, tools, commands
• Self-modification detection — monitors writes to extensions, AGENTS.md, .ssh, .env
• Session integrity — hash session files, detect reframe attacks

• sentinel_policy — View/modify permission policies
• sentinel_audit — Query audit trail
• sentinel_scan — Security scan for manipulation patterns

• /sentinel status — Current policies and audit stats
• /sentinel audit [n] — Last N audit entries
• /sentinel scan — Full security scan