What it does. Write custom Semgrep SAST rules in YAML to detect application-specific Part of mukul975/Anthropic-Cybersecurity-Skills (817 security skills) (mukul975/Anthropic-Cybersecurity-Skills).

Install

• npx skills add mukul975/Anthropic-Cybersecurity-Skills --skill implementing-semgrep-for-custom-sast-rules, or copy the skill folder into ~/.claude/skills/implementing-semgrep-for-custom-sast-rules/.
• Raw file: curl -sL https://raw.githubusercontent.com/mukul975/Anthropic-Cybersecurity-Skills/HEAD/skills/implementing-semgrep-for-custom-sast-rules/SKILL.md

SKILL.md (verbatim)

name: implementing-semgrep-for-custom-sast-rules
description: Write custom Semgrep SAST rules in YAML to detect application-specific
  vulnerabilities, enforce coding standards, and integrate into CI/CD pipelines.
domain: cybersecurity
subdomain: devsecops
tags:
- semgrep
- sast
- static-analysis
- custom-rules
- devsecops
- code-security
version: '1.0'
author: mahipal
license: Apache-2.0
nist_csf:
- PR.PS-01
- GV.SC-07
- ID.IM-04
- PR.PS-04
mitre_attack:
- T1195
- T1554
- T1059.004

Implementing Semgrep for Custom SAST Rules

Overview

Semgrep is an open-source static analysis tool that uses pattern-matching to find bugs, enforce code standards, and detect security vulnerabilities. Custom rules are written in YAML using Semgrep's pattern syntax, making it accessible without requiring compiler knowledge. It supports 30+ languages including Python, JavaScript, Go, Java, and C.

When to Use

• When deploying or configuring implementing semgrep for custom sast rules capabilities in your environment
• When establishing security controls aligned to compliance requirements
• When building or improving security architecture for this domain
• When conducting security assessments that require this implementation

Prerequisites

• Python 3.8+ or Docker
• Semgrep CLI installed
• Target codebase in a supported language

Installation

# Install via pip
pip install semgrep

# Install via Homebrew
brew install semgrep

# Run via Docker
docker run -v "${PWD}:/src" returntocorp/semgrep semgrep --config auto /src

# Verify
semgrep --version

Running Semgrep

# Auto-detect rules for your code
semgrep --config auto .

# Use Semgrep registry rules
semgrep --config r/python.lang.security

# Use custom rule file
semgrep --config my-rules.yaml .

# Use multiple configs
semgrep --config auto --config ./custom-rules/ .

# JSON output
semgrep --config auto --json . > results.json

# SARIF output for GitHub
semgrep --config auto --sarif . > results.sarif

# Filter by severity
semgrep --config auto --severity ERROR .

Writing Custom Rules

Basic Pattern Matching

# rules/sql-injection.yaml
rules:
  - id: sql-injection-string-format
    languages: [python]
    severity: ERROR
    message: |
      Potential SQL injection via string formatting.
      Use parameterized queries instead.
    pattern: |
      cursor.execute(f"..." % ...)
    metadata:
      cwe: ["CWE-89"]
      owasp: ["A03:2021"]
      category: security
    fix: |
      cursor.execute("SELECT * FROM users WHERE id = %s", (user_id,))

Pattern Operators

rules:
  - id: hardcoded-secret-in-code
    languages: [python, javascript, typescript]
    severity: ERROR
    message: Hardcoded secret detected in source code
    patterns:
      - pattern-either:
          - pattern: $VAR = "..."
          - pattern: $VAR = '...'
      - metavariable-regex:
          metavariable: $VAR
          regex: (?i)(password|secret|api_key|token|aws_secret)
      - pattern-not: $VAR = ""
      - pattern-not: $VAR = "changeme"
      - pattern-not: $VAR = "PLACEHOLDER"
    metadata:
      cwe: ["CWE-798"]
      category: security

Taint Analysis

rules:
  - id: xss-taint-tracking
    languages: [python]
    severity: ERROR
    message: User input flows to HTML response without sanitization
    mode: taint
    pattern-sources:
      - pattern: request.args.get(...)
      - pattern: request.form.get(...)
      - pattern: request.form[...]
    pattern-sinks:
      - pattern: return render_template_string(...)
      - pattern: Markup(...)
    pattern-sanitizers:
      - pattern: bleach.clean(...)
      - pattern: escape(...)
    metadata:
      cwe: ["CWE-79"]
      owasp: ["A03:2021"]

Multiple Language Rule

rules:
  - id: insecure-random
    languages: [python, javascript, go, java]
    severity: WARNING
    message: |
      Using insecure random number generator. Use cryptographically
      secure alternatives for security-sensitive operations.
    pattern-either:
      # Python
      - pattern: random.random()
      - pattern: random.randint(...)
      # JavaScript
      - pattern: Math.random()
      # Go
      - pattern: math/rand.Intn(...)
      # Java
      - pattern: new java.util.Random()
    metadata:
      cwe: ["CWE-330"]

Enforce Coding Standards

rules:
  - id: require-error-handling
    languages: [go]
    severity: WARNING
    message: Error return value not checked
    pattern: |
      $VAR, _ := $FUNC(...)
    fix: |
      $VAR, err := $FUNC(...)
      if err != nil {
        return fmt.Errorf("$FUNC failed: %w", err)
      }

  - id: no-console-log-in-production
    languages: [javascript, typescript]
    severity: WARNING
    message: Remove console.log before merging to production
    pattern: console.log(...)
    paths:
      exclude:
        - "tests/*"
        - "*.test.*"

JWT Security Rules

rules:
  - id: jwt-none-algorithm
    languages: [python]
    severity: ERROR
    message: JWT decoded without algorithm verification - allows token forgery
    patterns:
      - pattern: jwt.decode($TOKEN, ..., algorithms=["none"], ...)
    metadata:
      cwe: ["CWE-347"]

  - id: jwt-no-verification
    languages: [python]
    severity: ERROR
    message: JWT decoded with verification disabled
    patterns:
      - pattern: jwt.decode($TOKEN, ..., options={"verify_signature": False}, ...)
    metadata:
      cwe: ["CWE-345"]

Rule Testing

# rules/test-sql-injection.yaml
rules:
  - id: sql-injection-format-string
    languages: [python]
    severity: ERROR
    message: SQL injection via format string
    pattern: |
      cursor.execute(f"...{$VAR}...")

# Test annotation in test file:
# test-sql-injection.py
def bad_query(user_id):
    # ruleid: sql-injection-format-string
    cursor.execute(f"SELECT * FROM users WHERE id = {user_id}")

def good_query(user_id):
    # ok: sql-injection-format-string
    cursor.execute("SELECT * FROM users WHERE id = %s", (user_id,))

# Run rule tests
semgrep --test rules/

# Test specific rule
semgrep --config rules/sql-injection.yaml --test

CI/CD Integration

GitHub Actions

name: Semgrep SAST
on: [pull_request]

jobs:
  semgrep:
    runs-on: ubuntu-latest
    container:
      image: returntocorp/semgrep
    steps:
      - uses: actions/checkout@v4

      - name: Run Semgrep
        run: |
          semgrep --config auto \
            --config ./custom-rules/ \
            --sarif --output results.sarif \
            --severity ERROR \
            .

      - name: Upload SARIF
        uses: github/codeql-action/upload-sarif@v3
        with:
          sarif_file: results.sarif

GitLab CI

semgrep:
  stage: test
  image: returntocorp/semgrep
  script:
    - semgrep --config auto --config ./custom-rules/ --json --output semgrep.json .
  artifacts:
    reports:
      sast: semgrep.json

Configuration File

# .semgrep.yaml
rules:
  - id: my-org-rules
    # ... rules here

# .semgrepignore
tests/
node_modules/
vendor/
*.min.js

Best Practices

1. Start with auto config then add custom rules for org-specific patterns
2. Test rules with # ruleid: and # ok: annotations
3. Use taint mode for data flow vulnerabilities (XSS, SQLi, SSRF)
4. Include metadata (CWE, OWASP) for vulnerability classification
5. Provide fix suggestions with the fix key where possible
6. Exclude test files to reduce false positives
7. Version control rules in a shared repository
8. Run in CI as a blocking check for ERROR severity findings

Other files in this skill

• LICENSE
• references/api-reference.md
• references/standards.md
• scripts/agent.py

references/api-reference.md (verbatim)

API Reference: Semgrep Custom SAST Rules

Libraries Used

Installation

# Python package
pip install semgrep

# Homebrew (macOS)
brew install semgrep

# Docker
docker pull semgrep/semgrep:latest

CLI Reference

Core Commands

# Scan with auto-detected rules
semgrep scan --config auto --json --output results.json /path/to/code

# Scan with specific rulesets from Semgrep Registry
semgrep scan --config p/python --config p/owasp-top-ten /path/to/code

# Scan with a custom rule file
semgrep scan --config my-rules.yaml /path/to/code

# Scan with multiple configs
semgrep scan --config p/security-audit --config ./custom-rules/ /path/to/code

Key CLI Flags

Custom Rule Syntax

Basic Pattern Rule

rules:
  - id: hardcoded-password
    pattern: password = "..."
    message: "Hardcoded password detected — use environment variables"
    languages: [python]
    severity: ERROR
    metadata:
      cwe: ["CWE-798: Use of Hard-coded Credentials"]
      owasp: ["A07:2021 - Identification and Authentication Failures"]

Pattern Operators

rules:
  - id: sql-injection-format-string
    patterns:
      - pattern: |
          cursor.execute($QUERY % ...)
      - pattern-not: |
          cursor.execute("..." % ())
    message: "SQL injection via string formatting — use parameterized queries"
    languages: [python]
    severity: ERROR

  - id: unsafe-deserialization
    pattern-either:
      - pattern: pickle.loads(...)
      - pattern: pickle.load(...)
      - pattern: yaml.load(..., Loader=yaml.Loader)
      - pattern: yaml.unsafe_load(...)
    message: "Unsafe deserialization — may allow remote code execution"
    languages: [python]
    severity: ERROR

  - id: missing-timeout-requests
    patterns:
      - pattern: requests.$METHOD(...)
      - pattern-not: requests.$METHOD(..., timeout=..., ...)
    message: "HTTP request without timeout — may hang indefinitely"
    languages: [python]
    severity: WARNING

Metavariable Patterns

rules:
  - id: eval-user-input
    patterns:
      - pattern: |
          $INPUT = request.$METHOD(...)
          ...
          eval($INPUT)
    message: "User input passed to eval() — command injection risk"
    languages: [python]
    severity: ERROR

Python Integration

import subprocess
import json

def run_semgrep(target_path, config="auto", severity=None):
    cmd = [
        "semgrep", "scan",
        "--config", config,
        "--json",
        "--metrics", "off",
        str(target_path),
    ]
    if severity:
        cmd.extend(["--severity", severity])

    result = subprocess.run(cmd, capture_output=True, text=True, timeout=600)
    output = json.loads(result.stdout)
    return output.get("results", [])

def summarize_findings(results):
    by_severity = {"ERROR": [], "WARNING": [], "INFO": []}
    for r in results:
        sev = r.get("extra", {}).get("severity", "INFO")
        by_severity[sev].append({
            "rule": r["check_id"],
            "file": r["path"],
            "line": r["start"]["line"],
            "message": r["extra"]["message"],
        })
    return by_severity

Semgrep Registry Rule Packs

Output Format

{
  "results": [
    {
      "check_id": "python.lang.security.audit.eval-detected",
      "path": "app/views.py",
      "start": {"line": 42, "col": 5},
      "end": {"line": 42, "col": 28},
      "extra": {
        "message": "Detected eval() usage — avoid with untrusted input",
        "severity": "ERROR",
        "metadata": {
          "cwe": ["CWE-95"],
          "owasp": ["A03:2021 - Injection"]
        }
      }
    }
  ],
  "errors": [],
  "stats": {
    "findings": 3,
    "errors": 0,
    "total_time": 2.45
  }
}

references/standards.md (verbatim)

Standards - Semgrep Custom SAST Rules

OWASP Top 10 (2021) Coverage

CWE Coverage

Common CWEs detectable via Semgrep custom rules: CWE-79 (XSS), CWE-89 (SQLi), CWE-798 (Hardcoded Credentials), CWE-330 (Insecure Random), CWE-502 (Deserialization), CWE-918 (SSRF)

NIST SP 800-53 Rev 5

• SA-11: Developer Security Testing
• SA-15: Development Process, Standards, and Tools

Compliance

• PCI DSS v4.0 Req 6.3.2: Secure development with automated tools
• SOC 2 CC8.1: Change management with code scanning

Back to mukul975/Anthropic-Cybersecurity-Skills (817 security skills) or Agent skills.