CodeOptiX Integration

SuperClaw integrates with CodeOptiX for multi-modal security evaluation.

Overview

flowchart LR
    subgraph SuperClaw
        A[Behaviors]
        B[Attacks]
    end

    subgraph CodeOptiX
        C[BehaviorSpec]
        D[EvaluationEngine]
        E[StaticAnalyzer]
        F[LLMEvaluator]
    end

    A -->|Adapter| C
    C --> D
    D --> E
    D --> F

Components

SuperClawBehaviorAdapter

Bridges SuperClaw behaviors to CodeOptiX:

from superclaw.codeoptix import adapt_behavior_to_codeoptix

# Create adapted behavior
adapted = adapt_behavior_to_codeoptix("prompt-injection-resistance")

# Use with CodeOptiX
result = adapted.evaluate(agent_output)

SecurityEvaluator

Multi-modal evaluation:

from superclaw.codeoptix import create_security_evaluator

evaluator = create_security_evaluator(llm_provider="openai")

results = evaluator.evaluate(
    agent_output=output,
    behavior_names=["prompt-injection-resistance"],
)

# Results include:
# - Behavior evaluation
# - Static analysis
# - LLM judgment
# - Pattern matches

SecurityEvaluationEngine

Full evaluation workflow:

from superclaw.codeoptix import create_security_engine
from superclaw.adapters import create_adapter

adapter = create_adapter("openclaw", {"target": "ws://127.0.0.1:18789"})
engine = create_security_engine(adapter, llm_provider="anthropic")

result = engine.evaluate_security(
    behavior_names=["prompt-injection-resistance", "tool-policy-enforcement"]
)

print(f"Overall Score: {result.overall_score:.1%}")
print(f"Passed: {result.overall_passed}")

CLI Commands

# Check status
superclaw codeoptix status

# Register behaviors with CodeOptiX
superclaw codeoptix register

# Run multi-modal evaluation
superclaw codeoptix evaluate --target ws://127.0.0.1:18789 --llm-provider openai

Register with CodeOptiX

from superclaw.codeoptix import register_superclaw_behaviors

# Registers all SuperClaw behaviors as:
# - security-prompt-injection-resistance
# - security-tool-policy-enforcement
# - etc.
registered = register_superclaw_behaviors()

# Now use with CodeOptiX CLI
# codeoptix eval --behaviors security-prompt-injection-resistance

Multi-Modal Pipeline

sequenceDiagram
    participant E as Engine
    participant A as Adapter
    participant Agent
    participant S as StaticAnalyzer
    participant L as LLMEvaluator
    participant B as Behavior

    E->>A: send_prompt(attack)
    A->>Agent: Execute
    Agent-->>A: Response
    A-->>E: AgentOutput

    par Static Analysis
        E->>S: analyze(code)
        S-->>E: Issues
    and LLM Evaluation
        E->>L: evaluate(code)
        L-->>E: Judgment
    and Behavior Check
        E->>B: evaluate(output)
        B-->>E: Result
    end

    E->>E: Aggregate results