semgrep-rule-creator

Semgrep Rule Creator

When to Use

• Writing custom Semgrep rules for project-specific vulnerability patterns
• Creating detection rules for internal APIs or frameworks
• Building taint-mode rules to track data flow from sources to sinks
• Porting detection logic from other tools to Semgrep
• Defining coding standards as enforceable rules

When NOT to Use

• Running existing Semgrep scans (use the semgrep sub-skill in static-analysis)
• Complex interprocedural analysis (use CodeQL for deep data flow)
• Binary or bytecode analysis

Rule Template

rules:
  - id: <org>-<category>-<specific-issue>
    languages: [<language>]
    message: >
      <Clear description of the issue and remediation>
    severity: ERROR  # ERROR | WARNING | INFO
    metadata:
      cwe: "CWE-XXX: Description"
      confidence: HIGH  # HIGH | MEDIUM | LOW
      category: security
    patterns:
      - pattern: <vulnerable-code-pattern>
      - pattern-not: <safe-variant-to-exclude>

Pattern Operators

Operator	Use Case
pattern	Match exact code structure
patterns	All must match (AND)
pattern-either	Any matches (OR)
pattern-not	Exclude matches
pattern-inside	Only match within context
pattern-not-inside	Only match outside context
pattern-regex	Regex on source text
metavariable-regex	Regex on captured variable
metavariable-comparison	Numeric/string comparison

Taint Mode Template

rules:
  - id: <taint-rule-id>
    languages: [<language>]
    message: "Untrusted data flows to dangerous sink"
    severity: ERROR
    mode: taint
    pattern-sources:
      - pattern: <source-of-untrusted-data>
    pattern-sinks:
      - pattern: <dangerous-function-call>
    pattern-sanitizers:
      - pattern: <function-that-makes-input-safe>

Testing Rules

Create test files with annotations:

def test_vulnerable():
    # ruleid: my-rule-id
    dangerous_call(user_input)

def test_safe():
    # ok: my-rule-id
    safe_call(sanitized_input)

semgrep --test rules/

Best Practices

1. Use specific, descriptive rule IDs with org prefix
2. Include CWE and confidence metadata
3. Write clear remediation guidance in the message
4. Always create test cases for true positives AND false positives
5. Use pattern-not to reduce false positives
6. Prefer pattern-inside to limit scope rather than overly complex patterns
7. Test against real codebases before deploying to CI

Resources

• Rule Syntax — https://semgrep.dev/docs/writing-rules/rule-syntax/
• Playground — https://semgrep.dev/playground
• Rule Registry — https://semgrep.dev/explore