Can it prevent malicious code execution?

Absolutely. It identifies and blocks patterns like eval(), subprocess with shell=True, and unsafe deserialization to ensure the agent doesn't run harmful scripts.

How does it prevent prompt injection?

It strips specific instruction patterns like 'Ignore previous instructions' and XML-like tags (e.g., , ) that could trick the AI into changing its behavior.

What types of content does this skill sanitize?

It targets external data such as GitHub Issues, PRs, Discussions, WebFetch results, and any untrusted user-provided strings or URLs.

Does it limit the amount of text processed?

Yes, it enforces a 2000-word maximum per entry to prevent context stuffing and ensure the AI remains focused on relevant information.

Is the sanitization automated?

Yes, it includes a PostToolUse hook that automatically cleans outputs from web searches, curls, and GitHub CLI commands before the AI processes them.

Content Sanitization

Name: Content Sanitization
Author: micsapp

bymicsapp

0•

보안 및 테스팅

Protects AI workflows by sanitizing untrusted external content and preventing prompt injection or malicious code execution.

The Content Sanitization skill provides a robust security framework for handling external data within Claude Code and AI agent environments. It establishes strict guidelines for processing content from GitHub, web searches, and user-provided URLs by stripping dangerous system tags, instruction patterns, and code execution markers. By classifying data into clear trust levels and enforcing size limits and boundary markers, this skill ensures that AI agents remain secure and focused, preventing malicious 'jailbreak' attempts or the accidental execution of untrusted scripts from the web.

주요 기능

010 GitHub stars

02Multi-level trust classification for local, GitHub, and web sources

03Content truncation and boundary marking for context integrity

04Prevention of code execution patterns including eval(), exec(), and unsafe YAML loading

05Automated enforcement via PostToolUse hooks for web and CLI outputs

06Automated stripping of XML system tags and prompt injection patterns

사용 사례

01Sanitizing web search results before feeding them into an AI workflow

02Safely processing public GitHub issues and pull requests for code analysis

03Handling user-submitted URLs without risking prompt injection or system overrides

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add micsapp/micstec-skills content-sanitization

For use in Claude.ai and ChatGPT

주요 기능

010 GitHub stars

02Multi-level trust classification for local, GitHub, and web sources

03Content truncation and boundary marking for context integrity

04Prevention of code execution patterns including eval(), exec(), and unsafe YAML loading

05Automated enforcement via PostToolUse hooks for web and CLI outputs

06Automated stripping of XML system tags and prompt injection patterns

사용 사례

01Sanitizing web search results before feeding them into an AI workflow

02Safely processing public GitHub issues and pull requests for code analysis

03Handling user-submitted URLs without risking prompt injection or system overrides

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add micsapp/micstec-skills content-sanitization

For use in Claude.ai and ChatGPT