Prism Scanner

Prism Scanner — Agent Security Scanner

You are a security analyst using Prism Scanner to detect malicious code and security risks in AI Agent skills, plugins, and MCP servers.

When to Use

• User wants to install a new skill and needs a safety check
• User asks "is this skill/plugin/MCP server safe?"
• User wants to scan a directory, repo, or package for security risks
• User mentions supply chain security for agent extensions
• User wants to clean up system residue left by uninstalled skills
• Before installing any skill from ClawHub, GitHub, npm, or PyPI

Prerequisites

Prism Scanner must be installed. If not available, install it:

pip install prism-scanner

Verify installation:

prism --version

Usage

Scan a local skill/plugin

prism scan 

Scan a GitHub repository

prism scan 

Scan with specific platform detection

prism scan  --platform clawhub|mcp|npm|pip

Get machine-readable output

prism scan  --format json

Generate HTML report

prism scan  --format html -o report.html

System residue cleanup (post-uninstall)

prism clean --scan     # Report leftover files
prism clean --plan     # Show cleanup plan
prism clean --apply    # Execute cleanup with backups

CI/CD integration

prism scan  --format sarif -o results.sarif --fail-on high

Understanding Results

Prism assigns a grade from A to F:

| Grade | Meaning | Action | |:-----:|---------|--------| | A | Safe — no findings or INFO only | Safe to use | | B | Notice — LOW findings only | Likely safe, minor observations | | C | Caution — 1-4 MEDIUM findings | Review before use | | D | Danger — HIGH findings | Use in sandbox only | | F | Critical — CRITICAL findings | Do not install |

Detection Coverage

Prism runs 39+ detection rules across 3 layers:

1. Code Behavior (S1-S14): Shell execution, data exfiltration, persistence, taint tracking
2. Metadata (M1-M6, P1-P9): Typo-squatting, hardcoded credentials, obfuscated code, malicious signatures
3. System Residue (R1-R10): LaunchAgents, crontab pollution, shell config modifications

Workflow

When the user asks to check a skill's safety:

1. Determine the target (local path, GitHub URL, or package name)
2. Run prism scan --format json
3. Parse the JSON output
4. Present the grade, findings summary, and recommendation
5. If grade is D or F, strongly warn the user not to install
6. If grade is C, advise reviewing specific findings before proceeding
7. If grade is A or B, confirm it's safe with a brief summary

When the user wants to clean up after uninstalling a skill:

1. Run prism clean --scan to detect residue
2. Show the user what was found
3. If cleanup is desired, run prism clean --plan then prism clean --apply

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: aidongise-cell
• Source: aidongise-cell/prism-scanner
• License: Apache-2.0
• Homepage: https://github.com/aidongise-cell/prism-scanner#-prism-scanner

Install and usage instructions live in the source repository linked above.