Bastion AI
Security Tool for AI Agents

Supply-Chain Security
for AI Agents

Scan AI agent plugins and skills for security risks. Detect capability drift, surface hidden behaviors, and verify trust before deploying.

Deep static analysis across Python, JavaScript, and TypeScript — no runtime required.

Deploy Bastion
$ pip install bastion-ai-security
"I built Bastion to protect bots and AI agents as we head into a bot-first world. With platforms like OpenClaw and ai.com enabling autonomous agents at scale, the supply chain for AI skills and plugins needs the same security scrutiny we give to traditional software dependencies."
BB
Brantley Brumley
Real-World Incident — February 2026

The Most Downloaded ClawHub Skill Was Malware

The most popular skill on OpenClaw's ClawHub marketplace turned out to be an infostealer (Atomic Stealer/AMOS) disguised as a legitimate tool. Over 800 malicious skills were discovered, targeting crypto wallets, passwords, and credentials. No code review, no sandboxing, no security scanning was in place.

Bastion scans for exactly these threats — before they reach your agents.

Bastion Mascot

Your AI Agent's Plugins Could Be Doing Anything

AI agents rely on third-party plugins that can access your filesystem, make network requests, read secrets, and execute arbitrary code. Most teams have zero visibility into what these plugins actually do.

11
Capability categories scanned
0–100
Quantified trust scoring
4
Formats supported

Two Ways to Use Bastion

CLI for developers or MCP integration for autonomous AI agents

Path 1

Human-Operated CLI

You run scans, review results in the terminal, and make security decisions. Connect to the cloud dashboard for monitoring, trust scores, and verification.

$ pip install bastion-ai-security
$ bastion init
$ bastion scan
Path 2

AI Agent Integration (MCP)

Your AI agent uses Bastion as a security tool via MCP. It scans plugins, checks trust scores, and makes security decisions autonomously — no human in the loop.

$ pip install bastion-ai-security[mcp]
$ bastion mcp-serve
Agent calls scan_directory, check_trust_score...

Both paths produce the same security artifacts, trust scores, and detection coverage across all 11 capability categories.

Everything You Need to Secure AI Agents

Discovery, analysis, drift detection, trust scores, and CI/CD integration

Plugin Discovery

Auto-discovers plugins and skills across Python, JavaScript, and TypeScript projects. No manual configuration required.

Deep Static Analysis

AST-based analysis detecting 11 capability categories including dependency chain scanning across Python, JavaScript, TypeScript, and SKILL.md — from system execution to privilege escalation.

Drift Detection

Tracks changes between scans and alerts on new capabilities. Catch permission escalations before they become threats.

Remediation Guidance

Get actionable fix suggestions for every finding. Each detected capability includes specific remediation steps to secure your plugins.

Security Policy

Acknowledge known capabilities to reduce alert fatigue. Define a policy.yaml to track accepted risks without losing visibility.

Dependency Scanning

Analyze package.json, requirements.txt, and other dependency files for supply-chain risks including typosquats, URL installs, and unpinned versions.

Trust Scores

Quantified security scoring from 0–100. Portable trust scores that platforms and marketplaces can verify via API.

AI Bill of Materials

Generate machine-readable AIBOM documents. Complete security documentation for every plugin in your agent ecosystem.

CI/CD Integration

Non-zero exit codes on high/critical findings. Block unsafe deploys automatically in your existing pipeline.

MCP Server Integration

Use Bastion as an MCP server for AI agents. Scan directories, analyze files, and check trust scores directly from Claude, Cursor, or any MCP-compatible client.

OpenClaw Security

Purpose-built analysis for OpenClaw SKILL.md files. Detects dangerous bash commands, sandbox bypass flags, and external URL patterns in agent skill definitions.

What Bastion Detects

Ten capability categories analyzed through deep AST inspection

System Execution

subprocess, os.system, child_process

Network Access

HTTP requests, socket connections

Secret Access

Environment variables, credential files

Dynamic Code

eval, exec, Function constructor

Obfuscation

base64+exec, encoded payloads

File System Access

open, pathlib, fs module, glob

Sandbox Bypass

--yolo flags, elevated:true, sandbox escape

Install Risk

curl|bash, sudo, pip from URLs

Hardcoded Secrets

AWS keys, API tokens, passwords in source

Privilege Escalation

Combined exec + network + secrets

Up and Running in 4 Commands

From install to cloud-monitored security in under a minute

1

Install

pip install bastion-ai-security
2

Initialize

bastion init
3

Connect

bastion connect
4

Scan

bastion scan
terminal
$ bastion scan
Discovering plugins... found 5 plugins
Analyzing capabilities...
Plugin: email_sender.py
Risk: MEDIUM
Capabilities: network_access, secret_access
Plugin: data_processor.py
Risk: LOW
Capabilities: (none detected)
Plugin: web_scraper.py
Risk: HIGH
Capabilities: network_access, system_execution, dynamic_code, sandbox_bypass, hardcoded_secret
Plugin: formatter.py
Risk: LOW
Capabilities: (none detected)
Plugin: deploy-skill/SKILL.md
Risk: CRITICAL
Capabilities: sandbox_bypass, system_execution, install_risk
Scan complete: 5 plugins, 1 critical, 1 high, 1 medium, 2 low
Trust score: 64/100
Cloud Dashboard

Monitor Everything from One Place

Connect your local Bastion projects to the cloud for centralized security monitoring, historical tracking, and verification APIs.

  • Centralized monitoring across all your projects
  • Historical scan tracking and drift timeline
  • Trust score verification API for platforms and marketplaces
  • Real-time alerts on capability drift and risk changes
Sign In to Dashboard

Project Overview

All Verified
agent-toolkit
12 plugins · Last scan 2h ago
92
trust score
chat-plugins
8 plugins · Last scan 5h ago
74
trust score
automation-suite
6 plugins · Last scan 1d ago
98
trust score

Start Securing Your AI Agents

Open source, free to use, and ready in under a minute. Install locally or connect to the cloud dashboard.

$ pip install bastion-ai-security