skill-scanner

🥈Silver

Skill Scanner is a security tool for validating agent skills. It benefits operations teams by ensuring agent skills are secure before deployment. It integrates with Claude agents and is applied during the skill development and testing phases.

1,788120Updated 2w ago

Intermediate30min to implementautomation

Saves ~300 min per use

Quick InstallView Source

git clone https://github.com/cisco-ai-defense/skill-scanner.git

Works with:

Claude

Overview

About This Skill

How to Use

[{"step":"Install and configure Skill Scanner","action":"Run 'pip install skill-scanner' in your development environment. Ensure you have the latest version of the skill's manifest file and source code in your project directory.","tip":"Add Skill Scanner to your pre-commit hooks to automatically scan skills before each commit. Use 'skill-scanner --install-hooks' to set this up."},{"step":"Run the security scan","action":"Execute 'skill-scanner scan --skill-name [SKILL_NAME] --output-format json' in your terminal. For Claude agents, use 'claude skills scan [SKILL_NAME]' if integrated.","tip":"Run scans in both development and sandbox environments. Use 'skill-scanner scan --environment sandbox' to test skills in isolated conditions."},{"step":"Review and prioritize findings","action":"Examine the scan report for Critical and High severity issues first. Focus on hardcoded secrets, excessive permissions, and unsafe file operations as these pose the highest risk.","tip":"Use the 'skill-scanner report --severity critical --format detailed' command to get actionable remediation steps for each finding."},{"step":"Remediate vulnerabilities","action":"Implement the suggested fixes from the scan report. For hardcoded credentials, use environment variables or secret management systems. For file operations, add input validation and sanitization.","tip":"Test each fix by running the scan again. Use 'skill-scanner scan --compare-with-previous' to verify that issues have been resolved."},{"step":"Validate and document","action":"After remediation, run a final scan and document the security posture in your deployment checklist. Include the scan ID and timestamp in your deployment records.","tip":"For enterprise deployments, integrate Skill Scanner with your CI/CD pipeline using the 'skill-scanner --ci-mode' flag to automatically block deployments with security issues."}]

Setup & Installation

Quick Install

No install command available. Check the GitHub repository for manual installation instructions.

Alternative Install (Git Clone)

git clone https://github.com/cisco-ai-defense/skill-scanner

Requirements

Claude Code or compatible AI agent
Works with: Claude

Quick Start Guide

Install the Skill

Copy the install command above and run it in your terminal.

Open Your AI Agent

Launch Claude Code, Cursor, or your preferred AI coding agent.

Try It Out

Use the prompt template or examples below to test the skill.

Customize

Adapt the skill to your specific use case and workflow.

Usage Examples

Prompt Template

Scan the [SKILL_NAME] agent skill for security vulnerabilities using the Skill Scanner tool. Check for: 1) Unauthorized file system access patterns, 2) Suspicious network calls or external dependencies, 3) Hardcoded secrets or credentials, 4) Excessive permissions in the skill manifest. Generate a report with findings ranked by severity (Critical/High/Medium/Low) and provide remediation steps for each issue. Test the skill in a sandbox environment before deployment.

Example Output

```json
{
  "scan_id": "ss-20240518-001",
  "skill_name": "file-uploader-v2.1",
  "scan_timestamp": "2024-05-18T14:32:15Z",
  "findings": [
    {
      "severity": "Critical",
      "issue": "Hardcoded AWS credentials in skill manifest",
      "location": "skill.json (line 42)",
      "evidence": "AccessKeyId: AKIAIOSFODNN7EXAMPLE",
      "remediation": "Replace with environment variables. Use AWS IAM roles for production. Example: 'AWS_ACCESS_KEY_ID: ${env:AWS_ACCESS_KEY_ID}'",
      "status": "unresolved"
    },
    {
      "severity": "High",
      "issue": "Unrestricted file upload directory traversal",
      "location": "file_uploader.py (lines 89-95)",
      "evidence": "Path concatenation uses user input without sanitization: file_path = f"/uploads/{user_id}/{filename}"",
      "remediation": "Implement path sanitization using os.path.abspath() and validate file extensions. Add size limits (max 10MB).",
      "status": "unresolved"
    },
    {
      "severity": "Medium",
      "issue": "Excessive network permissions",
      "location": "skill.manifest.json",
      "evidence": "Network access granted to all domains: {\"network_access\": \"*\"}",
      "remediation": "Restrict to specific domains required for functionality. Example: {\"network_access\": [\"api.example.com\", \"files.example.com\"]}",
      "status": "resolved"
    }
  ],
  "summary": {
    "total_findings": 3,
    "critical_findings": 1,
    "high_findings": 1,
    "medium_findings": 1,
    "low_findings": 0,
    "recommended_actions": [
      "Address Critical and High severity issues before deployment",
      "Test patched skill in sandbox environment",
      "Re-scan after remediation to confirm fixes"
    ]
  }
}
```

**Security Scan Report for file-uploader-v2.1**

The Skill Scanner identified 3 security issues in the file-uploader-v2.1 agent skill during the pre-deployment validation phase. The most critical finding involves hardcoded AWS credentials that could allow unauthorized access to cloud storage if exposed. This risk is particularly severe as the credentials appear to be active (AKIAIOSFODNN7EXAMPLE is a known test key format). The file upload vulnerability could enable directory traversal attacks, potentially allowing attackers to write files outside the intended upload directory. While the network access issue has been marked as resolved in the manifest, the other two findings require immediate attention.

The scan was performed using Skill Scanner v1.2.3 against the skill's manifest and source code. The tool automatically checks for common security anti-patterns including hardcoded secrets, excessive permissions, and unsafe file operations. In production environments, similar scans should be integrated into the CI/CD pipeline to catch vulnerabilities before deployment.

For teams using Claude agents in enterprise environments, this tool provides critical validation that skills meet security baselines before they're deployed to production systems. The scan results include both technical details for developers and executive summaries for security teams, making it valuable for both technical and non-technical stakeholders.