Proxilion安全网关

Semantic Analysis (optional, requires Claude API)

ENABLE_SEMANTIC_ANALYSIS=false ANTHROPIC_API_KEY=sk-ant-xxx ```

Run performance benchmarks (requires criterion.rs)

cargo bench -p threat-engine

Run load tests (requires k6: brew install k6)

k6 run loadtest/baseline.js

Docker Deployment

```bash git clone https://github.com/clay-good/proxilion cd proxilion

Build from Source

```bash cargo build --release

TypeScript - install from local package

cd packages/mcp-proxilion-middleware && npm install && npm run build npm link # or copy to your project

Python - install from local package

cd packages/proxilion-mcp-python && pip install -e . ```

Deploy Behind Authentication

Proxilion does NOT authenticate users. Deploy architecture:

Client (with auth token)
    |
    v
API Gateway / Reverse Proxy (OAuth, API key validation)
    |
    v
Proxilion Gateway (threat analysis)
    |
    v
MCP Server (tool execution)

Build and run

cd tools && cargo build --release ./target/release/cost-calculator --requests 100000

Quick Start

Copy example and customize

cp proxilion-policy.example.toml proxilion-policy.toml

Integration Examples

Note: Client libraries are included in this repository but not yet published to npm/PyPI. Install from local source:

```bash

Run demo attack simulation

./demo.sh

Configuration

Environment Variables

```bash

Optional

LISTEN_ADDR=0.0.0.0:8787 ALERT_THRESHOLD=50 BLOCK_THRESHOLD=70 TERMINATE_THRESHOLD=90

Policy file (optional, for custom rules)

POLICY_FILE=./proxilion-policy.toml

API Reference

Components

Threat Analyzers (24 Active in Pipeline)

Pattern-Based (22): - Enumeration (nmap, masscan, port scanning) - Credential Access (.env, SSH keys, AWS credentials, /etc/shadow) - Exfiltration (curl to external IPs, pastebin uploads, netcat) - AI Velocity (automated execution patterns) - Prompt Engineering (jailbreak attempts) - Social Engineering (authority claims, roleplay manipulation) - Callback Validation (C2 detection, SSRF) - MCP Orchestration (malicious MCP chaining) - File Access (sensitive file patterns) - Task Fragmentation (command chaining, obfuscation) - Data Volume (bulk transfers, database dumps) - Privilege Escalation (sudo abuse, SUID) - Lateral Movement (SSH pivoting, internal network) - Hacking Tools (metasploit, hashcat, mimikatz) - Hallucination (AI fabrication detection) - Persistence (cron jobs, systemd, backdoors) - Defense Evasion (log clearing, firewall bypass) - Command and Control (reverse shells, Cobalt Strike) - Collection (data staging, screenshots) - Impact (destructive operations like rm -rf) - Tool Call (SQL injection, command chaining) - Legitimacy (context-aware false positive reduction)

Session-Aware (2): - Request Rate (burst detection, machine-like velocity) - Session Progression (kill chain tracking: Recon -> Credentials -> Exfiltration)

Additional Analyzers (6, require explicit configuration): - AI Autonomy (autonomous agent detection) - Conversation Analysis (social engineering via conversation context) - Conversation Context (multi-turn tracking) - Multi-Target Orchestration (parallel target operations) - Multi-User Correlation (coordinated attacks across users) - Semantic Analysis (Claude API-based intent analysis, optional feature flag)

---

Show comparison table

./target/release/cost-calculator --compare

Example output:

TOTAL MONTHLY COST: $64.06 Cost per request: $0.000641 ```

---