AgentSkillsCN

security-guidance

遵循安全最佳实践，防范安全漏洞。在编写处理用户输入、外部数据、Shell命令或敏感操作的代码时使用此功能。自动检查常见的安全反模式。

SKILL.md
--- frontmatter
name: security-guidance
description: Security best practices and vulnerability prevention. Use when writing code that handles user input, external data, shell commands, or sensitive operations. Automatically checks for common security anti-patterns.

Security Guidance

This skill provides security awareness when writing code. Be vigilant about these common vulnerability patterns.

Critical Security Patterns to Avoid

1. Command Injection

Never use untrusted input in shell commands:

javascript
// DANGEROUS - command injection vulnerability
exec(`command ${userInput}`)
child_process.exec(`ls ${directory}`)

// SAFE - use execFile with array arguments
execFile('command', [userInput])
spawn('ls', [directory])
python
# DANGEROUS
os.system(f"command {user_input}")

# SAFE
subprocess.run(['command', user_input], check=True)

2. GitHub Actions Workflow Injection

When editing .github/workflows/*.yml files:

yaml
# DANGEROUS - attacker can inject commands via PR title
run: echo "${{ github.event.pull_request.title }}"

# SAFE - use environment variables
env:
  TITLE: ${{ github.event.pull_request.title }}
run: echo "$TITLE"

Risky GitHub context variables:

  • •github.event.issue.title / body
  • •github.event.pull_request.title / body
  • •github.event.comment.body
  • •github.event.commits.*.message
  • •github.event.head_commit.message
  • •github.head_ref

3. XSS (Cross-Site Scripting)

javascript
// DANGEROUS
element.innerHTML = userContent
document.write(userContent)
dangerouslySetInnerHTML={{ __html: userContent }}

// SAFE
element.textContent = userContent  // For plain text
DOMPurify.sanitize(userContent)    // If HTML needed

4. Code Injection

javascript
// DANGEROUS - arbitrary code execution
eval(userInput)
new Function(userInput)

// SAFE - use specific parsers
JSON.parse(jsonString)  // For JSON data

5. Unsafe Deserialization

python
# DANGEROUS - pickle can execute arbitrary code
pickle.loads(untrusted_data)

# SAFE - use JSON or other safe formats
json.loads(untrusted_data)

Security Checklist

When writing code, verify:

  • • User input is validated before use
  • • Shell commands use array arguments, not string interpolation
  • • HTML content is sanitized before insertion
  • • No eval(), new Function(), or equivalent with user data
  • • Secrets are not hardcoded or logged
  • • SQL queries use parameterized statements
  • • File paths are validated to prevent path traversal
  • • Authentication/authorization checks are in place
  • • Error messages don't expose sensitive information
  • • Dependencies are from trusted sources

When to Be Extra Careful

  • •Editing workflow files (.github/workflows/)
  • •Handling file uploads
  • •Processing user-generated content
  • •Building shell commands
  • •Deserializing data
  • •Rendering HTML from external sources
  • •Working with authentication/authorization
  • •Accessing databases
  • •Making HTTP requests with user-controlled URLs

Resources