v2.2.0 — Sub-Agent Security + Security Dashboard

Phase 1: Sub-Agent Security

~510 LOC across 9 steps. 33 new tests passing.

Agent Trust & Access Control

• Agent Trust Scoring — hierarchy encoding with > separator, 0.7× decay per level, max depth 5
• Memory Access Control — read ACLs (trust tiers: ≥0.7 full, 0.5–0.7 shared, <0.5 own only), write restrictions, rate limiting (20/min), auto-quarantine
• Credential Isolation — RESTRICTED memories blocked below trust 0.7
• Quarantine System — auto-expiry (7-day TTL), batch approve by source
• Uninstall Protection — blocks uninstall from agent contexts

Environment-Based Source Inference

• Auto-detects Claude Code agents from process env vars (not forgeable via MCP params)
• Priority: SHIELDCORTEX_AGENT_SOURCE > CLAUDE_CODE_ENTRYPOINT > CLAUDE_AGENT_CONTEXT
• Missing-source audit logging for visibility into unconfigured callers
• strictSourceMode config: unknown sources get trust 0.3 and quarantined writes

Security Dashboard

• Security-first redesign: Shield | Audit | Quarantine | Memories | Brain | Graph
• Firewall mode read-only in dashboard (CLI-only changes)
• Threat timeline with severity indicators

Bug Fixes

• Defence pipeline now runs on MCP remember calls (was skipped when source undefined)
• Project filter for audit/quarantine tables
• Quarantine API schema mapping fix

Full Changelog: Drakon-Systems-Ltd/ShieldCortex@v2.1.4...v2.2.0