Rules Reference

Built-in Rules

AI Guard includes 5 built-in rules that detect common AI-generated code patterns:

1. excessive-comments

Detects files where the ratio of comment lines to total lines exceeds the configured threshold.

Property	Value
Rule Name	excessive-comments
Default Threshold	15% (0.15)
Configurable	Yes (maxCommentsRatio)
Severity	error

Example violation:

typescript

// This function processes user data
// It takes a user object as input
// Then it validates the data
// After validation it returns the result
function processUser(user) {
  // First we check if user exists
  if (!user) {
    // No user provided
    return null;
  }
  // Continue processing...
}

// Line 1: [excessive-comments] Comment ratio 45% exceeds threshold of 15%

How to fix: Remove redundant comments and let code speak for itself.

2. obvious-comments

Flags comments that simply restate what the code does, adding no value.

Property	Value
Rule Name	obvious-comments
Pattern Detection	AST-based + heuristics
Configurable	No
Severity	error

Common obvious comment patterns:

typescript

// Loop through users
for (const user of users) {
  // Active users only
  if (user.isActive) {
    // Add to active list
    activeUsers.push(user);
  }
}

// All three comments are obvious - the code is self-explanatory

How to fix: Remove comments that don't add information beyond what's in the code.

3. generic-names

Detects use of generic variable and function names that don't convey meaning.

Property	Value
Rule Name	generic-names
Default List	data, result, item, temp, value, obj, arr, info, content, response
Configurable	Yes (forbidGenericNames)
Severity	error

Example violations:

typescript

// Bad: generic names
async function fetchData() {
  const response = await api.get("/users");
  const data = response.data;
  const result = processData(data);
  return result;
}

// Good: descriptive names
async function fetchUsers() {
  const apiResponse = await api.get("/users");
  const users = apiResponse.data;
  const validatedUsers = validateUsers(users);
  return validatedUsers;
}

How to fix: Use domain-specific, descriptive names that convey the data's purpose.

4. large-functions

Flags functions that exceed the maximum line count threshold.

Property	Value
Rule Name	large-functions
Default Threshold	60 lines
Configurable	Yes (maxFunctionLines)
Severity	error

Example violation:

typescript

function processOrder(order) {
  // 80 lines of code...
  // Multiple responsibilities
  // Hard to test and maintain
}

// Line 1: [large-functions] Function 'processOrder' has 80 lines (max: 60)

How to fix: Break large functions into smaller, focused functions with single responsibilities.

5. ai-patterns

Detects phrases commonly used by AI code generators that indicate machine-generated content.

Property	Value
Rule Name	ai-patterns
Default Patterns	this function, we will, we need to, let's, in order to, going to, simply, just
Configurable	Yes (aiPatterns.forbid)
Severity	error

Common AI phrase examples:

typescript

// This function will process the user data
function processUserData(user) {
  // First, we need to validate the input
  if (!user) {
    return null;
  }

  // Next step: transform into output format
  const transformed = transform(user);

  // Finally, in order to complete the process, we return the result
  return transformed;
}

// Multiple AI-typical phrases detected

How to fix: Rewrite comments in a more natural, human style, or remove them if they're not needed.

Configuring Rules

Adjust rule thresholds in your configuration file:

typescript

export default defineConfig({
  // Allow more comments in documentation-heavy projects
  maxCommentsRatio: 0.20,

  // Stricter function length for microservices
  maxFunctionLines: 40,

  // Add project-specific generic names to avoid
  forbidGenericNames: [
    'data',
    'result',
    'item',
    'temp',
    'value',
    'myVar',  // custom
    'helper', // custom
  ],

  // Add custom AI patterns specific to your LLM
  aiPatterns: {
    forbid: [
      'this function',
      'we will',
      'we need to',
      'as mentioned above', // custom
      'it is worth noting',  // custom
    ],
  },
});

All rules run on every analysis. There is currently no way to disable individual rules, but you can adjust thresholds to be more or less strict.

Programmatic API