tools-only Phase 1 Checkpoint 3: Tier 2 Security Frameworks Analysis
Date: January 25, 2026 Status: Complete
Security Frameworks Analyzed
1. TikiTribe/claude-secure-coding-rules
Purpose: Open-source security rules guiding Claude Code to generate secure code by default
Standards Coverage:
- OWASP Top 10 2025
- AI/ML Security (NIST AI RMF, MITRE ATLAS, Google SAIF)
- Agentic AI Security
- 100+ rule sets
Repository Structure:
_core/ owasp-2025.md, ai-security.md, agent-security.md, rag-security.md
languages/ 12 languages (Python, JavaScript, TypeScript, Go, Rust, Java, C#, Ruby, R, C++, Julia, SQL)
backend/ 5 frameworks + 11 AI/ML (FastAPI, Express, Django, Flask, NestJS, LangChain, CrewAI, AutoGen, etc.)
rag/ 51 tools (orchestration, vector DBs, embeddings, chunking, search/rerank)
frontend/ 5 frameworks (React, Next.js, Vue, Angular, Svelte)
iac/ Terraform, Pulumi
containers/ Docker, Kubernetes
cicd/ GitHub Actions, GitLab CI
Rule Format:
DO: Specific secure implementation
DON'T: Specific insecure pattern
WHY: Explanation of risk
REFS: Links to authoritative sources
Enforcement Levels:
| Level | Behavior |
|---|---|
| strict | Refuse to generate insecure code |
| warning | Warn and offer secure alternatives |
| advisory | Mention security consideration |
Hierarchy: Global (~/.claude/CLAUDE.md) → Project (.claude/CLAUDE.md) → Directory (src/.claude/CLAUDE.md)
ZERG Application: Comprehensive security rule library for worker context, hierarchical rule application pattern
2. fr0gger/nova-claude-code-protector
Purpose: Security monitoring and prompt injection defense using NOVA Framework
Architecture: Claude Code hook system (SessionStart, PreToolUse, PostToolUse, SessionEnd)
Protection Modes:
| Mode | Trigger | Action |
|---|---|---|
| ACTIVE | PreToolUse | Blocks dangerous commands before execution |
| PASSIVE | PostToolUse | Warns Claude after content read |
Blocked Commands (ACTIVE): rm -rf, sudo rm, mkfs, dd, fork bombs, credential exfiltration
Three-Tier Detection:
| Tier | Latency | Method |
|---|---|---|
| Keywords | ~1ms | Regex patterns for known attacks |
| Semantics | ~50ms | ML similarity for paraphrased attacks |
| LLM | ~500-2000ms | AI evaluation for sophisticated attacks |
Attack Categories: Instruction Override, Jailbreak/Role-Playing, Encoding/Obfuscation, Context Manipulation
Session Tracking: JSONL logs, HTML reports (timeline, filtering), AI-powered summaries
Custom Rules (.nov files): meta, keywords, semantics, llm, condition sections
ZERG Application: Hook architecture for orchestrator monitoring, session logging for audit trail, three-tier detection for worker outputs
3. github/codeql-coding-standards
Purpose: CodeQL queries for safety-critical C/C++ compliance
Standards Coverage: AUTOSAR C++14, SEI CERT C++, SEI CERT C, MISRA C 2012
Key Features:
- Deviation records (permitted rule violations)
- Guideline re-categorization plans
- SARIF 2.1.0 output for compliance reporting
- Query suites for different rule sets
ZERG Application: Compliance management architecture, deviation record pattern
Security Pattern Synthesis
Defense-in-Depth Architecture
┌─────────────────────────────────────────────────────────────┐
│ LAYER 1: DESIGN TIME │
│ claude-secure-coding-rules: Generate secure code by default│
└─────────────────────────────────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────────┐
│ LAYER 2: RUNTIME │
│ nova-protector: Monitor and block dangerous operations │
└─────────────────────────────────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────────┐
│ LAYER 3: ANALYSIS │
│ codeql-coding-standards: Static analysis for compliance │
└─────────────────────────────────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────────┐
│ LAYER 4: AUDIT │
│ Session reporting and compliance documentation │
└─────────────────────────────────────────────────────────────┘
Critical Gaps in ZERG
| Gap | Impact | Mitigation Source |
|---|---|---|
| No secure code generation rules | Workers may generate vulnerable code | claude-secure-coding-rules |
| No runtime protection | Prompt injection attacks undetected | nova-protector |
| No audit trail | Cannot verify what workers did | nova-protector session tracking |
| No compliance management | Cannot document deviations | codeql-coding-standards patterns |
Recommended Security Integration
Phase 1: Foundation
- Add claude-secure-coding-rules to worker context
- Implement hook architecture for monitoring
Phase 2: Runtime Protection
- Integrate nova-protector patterns for orchestrator
- Add session logging for audit trail
- Implement three-tier detection for worker outputs
Phase 3: Compliance
- Add deviation record support for task specifications
- Implement SARIF output for analysis integration