CODITECT Agent Security Layer
AI agent security patterns, risk scoring, and supply chain trust verification for the CODITECT platform.
Overview
The Agent Security Layer protects CODITECT's 776 AI agents from:
- Prompt injection via tool inputs
- Secret exfiltration via tool outputs
- Destructive commands through Bash execution
- PII leakage in agent communications
- Supply chain attacks via malicious skills/plugins
Architecture
Built as native CODITECT hooks (PreToolUse / PostToolUse / PreAgentStart) with:
- SecurityGateHook — Intercepts tool calls before execution
- PatternEngine — YAML-based pattern matching with 25+ rules
- RiskAnalyzer — Hybrid 0-100 scoring with severity classification
- ActionRouter — Severity-based action dispatch (block/redact/confirm/warn/log)
Research Origin
Derived from the ClawGuard ecosystem research:
- ClawGuardian (superglue-ai) — Hook architecture, severity-action routing
- clawguard (JaydenBeard) — 55+ risk patterns, kill switch, multi-gateway
- ClawGuard (maxxie114) — Risk scoring algorithm, sanitization patterns
All source repositories are MIT licensed. This implementation is CODITECT-native (no OpenClaw dependency).
Quick Start
# Install dependencies
pip install -e ".[dev]"
# Run tests
pytest tests/
# View security patterns
ls config/security-patterns/
Project Status
| Track | Domain | Progress |
|---|---|---|
| R | Research & Analysis | 100% |
| D | Development & Engineering | 0% |
| T | Testing & Quality | 0% |
| O | Operations & Compliance | 0% |
See internal/project/plans/tracks/ for detailed task tracking.
Key Decisions (ADRs)
| ADR | Decision |
|---|---|
| ADR-001 | Build CODITECT-native, no OpenClaw dependency |
| ADR-002 | YAML pattern library format |
| ADR-003 | Fail-open default, fail-closed for autonomous loops |
| ADR-004 | Hybrid numeric + categorical risk scoring |
| ADR-005 | Three-layer supply chain defense |
License
Proprietary — AZ1.AI INC