SecureClaw v2.1.0 — Multi-Framework Security Coverage
5 frameworks. 15 rules. 55 checks. Kill switch. Behavioral baselines.
What's New
3 new behavioral rules:
- Rule 13 — Memory trust: block untrusted content from entering cognitive files
- Rule 14 — Kill switch: instant agent suspension via file flag
- Rule 15 — Reasoning telemetry: agents must state plans before multi-step operations
3 new CLI commands:
- secureclaw kill / secureclaw resume — emergency agent suspension
- secureclaw baseline — behavioral tool-call frequency tracking
4 new audit checks:
- SC-TRUST-001 — injection detection in cognitive files (MITRE AML.CS0051)
- SC-KILL-001 — kill switch status
- SC-CTRL-001 — default control token spoofing risk
- SC-DEGRAD-001 — missing graceful degradation config
2 new config options:
- failureMode — block_all / safe_mode / read_only
- riskProfile — strict / standard / permissive
Framework Coverage
| Framework | Score |
|---|---|
| OWASP ASI Top 10 | 10/10 |
| MITRE ATLAS Agentic | 10/14 |
| MITRE ATLAS OpenClaw | 4/4 cases |
| CoSAI Principles | 13/18 |
| CSA Singapore | 8/11 |
New Docs
- docs/openclaw-threat-model.md — 8 threat classes, 10 Mermaid diagrams
- docs/openclaw-attack-examples.md — 8 real attack scenarios with detection breakdowns