VANT v0.8.4 - Security & Feature Release
Major release with comprehensive security hardening and new features!
🔐 Security Fixes (12 Vulnerabilities)
| ID | Severity | Vector | Fix |
|---|---|---|---|
| V001 | CRITICAL | Command injection | Safe exec |
| V002 | HIGH | Token exposure | Remove from URL |
| V003 | HIGH | Auth bypass | MCP auth |
| V004-V009 | MEDIUM | DoS | Safe JSON parsing |
| V010 | HIGH | Prompt injection | 17+ patterns |
| V011 | MEDIUM | Key injection | Validation |
| V012 | LOW | Context overflow | Limits |
🔒 AI Security Hardening
- Prompt Injection: 17+ patterns blocked
- "ignore previous instructions", "forget everything"
- "new system:", "role:", "act as"
- "DAN mode", "jailbreak"
- Template injection: {{system}}, [INST], [SYS]
✨ New Features
- Entropy-Patch Protocol: Compress brain updates
vant compress- Entropy-Patch encoder- Optimized for memory efficiency
- Enhanced CLI: 25+ commands
vant onboard- Onboarding flowvant succession- Succession managementvant update- Update checkervant rate- GitHub rate limit
- Better Default Brain
- Improved identity, goals, lessons
- 24 model files in public brain
- MCP Server: Model Context Protocol support
- Steganography: Password-protected PNG encoding
- Weekly Security Audits: Automated
📚 Documentation
- All 42 docs version-aligned
- Manual brain creation guide
- Release process guide
- AGENTS.md with full policies
- CLI reference
✅ What's New
- Version: v0.8.4
- Release branches: release/v0.8.2, release/v0.8.4
- Tests: 15/15 passing
Thanks
Security audit and feature development by OpenHands agent.
VANT - Versatile Autonomous Networked Tool