Skip to main content
Automatically warns about potential security vulnerabilities when Claude edits files, helping prevent common security mistakes.

​
Overview

The Security Guidance plugin monitors file edits for common security patterns and provides warnings before Claude writes potentially vulnerable code. It acts as a safety net, catching security issues early.

​
Installation

This plugin is bundled with Claude Code. To enable it: 
/plugin enable security-guidance

​
How It Works

The plugin uses a PreToolUse hook that:
  1. Monitors all file edit operations
  2. Checks file paths and content for security patterns
  3. Shows warnings when risky patterns are detected
  4. Provides specific guidance for each vulnerability type
The hook runs before Claude writes the file, giving you a chance to reconsider the approach.

​
Security Patterns Detected

The plugin monitors for 9 common security issues:

​
1. GitHub Actions Command Injection

Files: .github/workflows/*.yml, .github/workflows/*.yaml What it catches:
  • Untrusted input used directly in run: commands
  • Issue titles, PR descriptions, commit messages in shell commands
Guidance provided: 
# UNSAFE ❌
run: echo "${{ github.event.issue.title }}"

# SAFE ✅  
env:
  TITLE: ${{ github.event.issue.title }}
run: echo "$TITLE"

​
2. SQL Injection

Files: *.py, *.js, *.ts, *.php, *.rb What it catches:
  • String concatenation in SQL queries
  • Unparameterized queries with user input
Guidance provided: 
# UNSAFE ❌
query = f"SELECT * FROM users WHERE id = {user_id}"

# SAFE ✅
query = "SELECT * FROM users WHERE id = ?"
cursor.execute(query, (user_id,))

​
3. Command Injection

Files: *.py, *.js, *.ts, *.sh, *.bash What it catches:
  • Using os.system(), eval(), exec() with user input
  • Shell commands with unvalidated strings
Guidance provided: 
# UNSAFE ❌
os.system(f"convert {filename} output.png")

# SAFE ✅
subprocess.run(["convert", filename, "output.png"], check=True)

​
4. Cross-Site Scripting (XSS)

Files: *.jsx, *.tsx, *.vue, *.html What it catches:
  • dangerouslySetInnerHTML usage
  • innerHTML assignments
  • Unescaped user content in templates
Guidance provided: 
// UNSAFE ❌
<div dangerouslySetInnerHTML={{__html: userContent}} />

// SAFE ✅
<div>{DOMPurify.sanitize(userContent)}</div>

​
5. Path Traversal

Files: All file types with file path operations What it catches:
  • User-controlled file paths
  • Missing validation on path strings
  • Direct use of ../ patterns
Guidance provided: 
// UNSAFE ❌
fs.readFile(`/uploads/${userFilename}`);

// SAFE ✅
const safe = path.basename(userFilename);
fs.readFile(path.join('/uploads', safe));

​
6. Insecure Deserialization

Files: *.py, *.js, *.java What it catches:
  • Using pickle.load() on untrusted data
  • Unsafe JSON.parse() patterns
  • Java deserialization of external data
Guidance provided: 
# UNSAFE ❌
data = pickle.load(user_file)

# SAFE ✅
import json
data = json.loads(user_file.read())

​
7. Hardcoded Secrets

Files: All code files What it catches:
  • API keys in source code
  • Passwords in configuration
  • Tokens and credentials hardcoded
Guidance provided: 
# UNSAFE ❌
API_KEY = "sk_live_abc123..."

# SAFE ✅  
API_KEY = os.environ.get("API_KEY")

​
8. Insecure Randomness

Files: *.py, *.js, *.java, *.go What it catches:
  • Using Math.random() for security tokens
  • Non-cryptographic random for secrets
Guidance provided: 
// UNSAFE ❌
const token = Math.random().toString(36);

// SAFE ✅
const token = crypto.randomBytes(32).toString('hex');

​
9. Unsafe HTML Rendering

Files: *.jsx, *.tsx, *.vue, *.svelte What it catches:
  • Direct HTML string rendering
  • Unsafe component props
Guidance provided: 
// UNSAFE ❌
<Component html={userInput} />

// SAFE ✅
<Component text={DOMPurify.sanitize(userInput)} />

​
Example Warnings

When Claude attempts to edit a GitHub Actions workflow: 
⚠️  SECURITY REMINDER ⚠️

You are editing a GitHub Actions workflow file.
Be aware of these security risks:

1. Command Injection: Never use untrusted input (like issue
   titles, PR descriptions) directly in run: commands
   
2. Use environment variables: Instead of 
   ${{ github.event.issue.title }}, use env: with proper
   quoting
   
3. Review the guide:
   https://github.blog/security/vulnerability-research/
   how-to-catch-github-actions-workflow-injections

Example of UNSAFE pattern to avoid:
run: echo "${{ github.event.issue.title }}"

Example of SAFE pattern:
env:
  TITLE: ${{ github.event.issue.title }}
run: echo "$TITLE"

Other risky inputs: pull_request.title, issue.body,
comment.body, head_commit.message

​
Configuration

No configuration required. The plugin activates automatically when enabled.

​
Session-Based Warnings

Warnings are shown once per session per file. This prevents:
  • Repetitive warnings on the same file
  • Disruption to workflow
  • Warning fatigue

​
When to Disable

Consider disabling if:
  • Working on non-security-sensitive code
  • Warnings become disruptive
  • You have other security tooling (e.g., CodeQL)
/plugin disable security-guidance

​
Technical Details

​
Hook Implementation

The plugin uses Python for pattern matching: 
# From hooks/security_reminder_hook.py
SECURITY_PATTERNS = [
  {
    "ruleName": "github_actions_workflow",
    "path_check": lambda path: ".github/workflows/" in path,
    "reminder": "..."
  },
  # ... 8 more patterns
]

​
State Management

Warnings are tracked per session:
  • State file: /tmp/security-warnings-{session-id}.json
  • Cleared when session ends
  • Survives session resume

​
Limitations

This plugin provides guidance, not absolute protection. Always:
  • Review security-sensitive code carefully
  • Use static analysis tools (SAST)
  • Perform security testing
  • Follow your organization’s security policies
The plugin:
  • Cannot detect all vulnerabilities
  • Uses simple pattern matching
  • May have false negatives
  • Should supplement, not replace, security reviews

​
Details

Name: security-guidanceType: PreToolUse Hook (Python)Author: David Dworken ([email protected])Version: 1.0.0Hook File: hooks/security_reminder_hook.pyPatterns Monitored: 9 security vulnerability types

Hook Development

Build your own PreToolUse hooks

Project Config

CLAUDE.md security guidelines

Build docs developers (and LLMs) love

Get started for freeTalk to us