ai-code-security-scanner

v1.0.0

Published

22 days ago

AST-based security scanner for AI-generated code

0High
0Medium
0Low

AI Code Security Scanner

AST-based security scanner for AI-generated code. Detects 43+ vulnerability types across 14 languages with auto-fix, 7 output formats, VS Code integration, and CI/CD support.

Try it instantly

npx ai-sec-scan ./src

Quickstart

# Install
npm install -g ai-code-security-scanner

# Scan a directory
ai-sec-scan ./src

# Scan and output JSON
ai-sec-scan ./src --format json

# Apply safe auto-fixes
ai-sec-scan ./src --fix

Supported Languages

| Language | Detection | Auto-fix | |----------|-----------|----------| | TypeScript / JavaScript | ✓ | ✓ | | Python | ✓ | ✓ | | Go | ✓ | ✓ | | Java | ✓ | ✓ | | C# | ✓ | ✓ | | Ruby | ✓ | ✓ | | PHP | ✓ | ✓ | | Rust | ✓ | ✓ | | Swift | ✓ | ✓ | | Kotlin | ✓ | ✓ | | C / C++ | ✓ | ✓ |

CI Integration

# .github/workflows/security-scan.yml
name: Security Scan
on: [push, pull_request]
jobs:
  scan:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: actions/setup-node@v4
        with:
          node-version: '20'
      - run: npx ai-sec-scan ./src --format sarif --min-severity high
        continue-on-error: true
      - uses: github/codeql-action/upload-sarif@v3
        with:
          sarif_file: output.sarif

Usage

CLI

# Scan a file
npx ai-sec-scan src/app.ts

# Scan a directory
npx ai-sec-scan ./src

# JSON output
npx ai-sec-scan ./src --json

# SARIF 2.1.0 output (for GitHub Security tab / CI artifacts)
npx ai-sec-scan ./src --sarif

# Filter by minimum severity to report
npx ai-sec-scan ./src --severity high

# Set minimum severity that triggers a non-zero exit code
npx ai-sec-scan ./src --min-severity critical

# Exclude paths matching a glob (repeatable)
npx ai-sec-scan . --ignore '**/node_modules/**' --ignore 'dist/**' --ignore '**/*.test.ts'

# Use a config file
npx ai-sec-scan ./src --config .ai-sec-scan.json

# Watch mode — re-scans on file changes, prints a diff of new/resolved findings
npx ai-sec-scan ./src --watch

CLI flags

| Flag | Description | |------|-------------| | [path] | File or directory to scan. Defaults to . | | --json | Output as JSON | | --sarif | Output as SARIF 2.1.0 | | --format <text\|json\|sarif> | Explicit format selector (overrides --json / --sarif) | | --severity <level> | Minimum severity to include in output (critical\|high\|medium\|low). Default: low | | --min-severity <level> | Severity that triggers a non-zero exit code. Default: high | | --ignore <glob> | Exclude matching paths (repeatable) | | --config <path> | Path to a .ai-sec-scan.json config file | | --watch | Watch for file changes and print a live diff of findings |

Config file (`.ai-sec-scan.json`)

Place a .ai-sec-scan.json in your project root (or pass --config):

{
  "severity": "medium",
  "format": "sarif",
  "fix": true,
  "ignore": ["dist/**", "**/*.test.ts", "**/*.spec.ts"]
}

| Key | Type | Description | |-----|------|-------------| | severity | string | Minimum severity to include (critical | high | medium | low). Default: low | | format | string | Default output format (text | json | sarif | html | junit). Default: text | | fix | boolean | Apply auto-fixes for supported finding types (equivalent to --fix on every run). Default: false | | ignore | string[] | Glob patterns to exclude from scanning (merged with --ignore flags). |

CLI flags override config file values.

Ignore file (`.aiscanner`)

Create a .aiscanner file in your project root (gitignore-style). Each non-comment, non-empty line is a glob pattern:

# ignore generated files
dist/**
coverage/**
**/*.min.js

Patterns from .aiscanner are merged with --ignore flags (CLI takes precedence).

API Server

npm run dev:server   # starts on http://localhost:3001

# Health check
curl http://localhost:3001/health

# Scan code snippet
curl -X POST http://localhost:3001/scan \
  -H "Content-Type: application/json" \
  -d '{"code": "const password = \"hunter2\""}'

# With AI explanations (requires ANTHROPIC_API_KEY)
curl -X POST http://localhost:3001/scan \
  -H "Content-Type: application/json" \
  -d '{"code": "eval(userInput);", "aiExplain": true}'

API authentication

When SERVER_API_KEY is set, every request (except GET /health) must include:

Authorization: Bearer <your-server-api-key>

export SERVER_API_KEY=my-secret-key

curl -X POST http://localhost:3001/scan \
  -H "Authorization: Bearer $SERVER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"code": "const pw = \"hunter2\""}'

Requests without a valid token receive 401 Unauthorized.

Web UI

cd web && npm install && npm run dev

Open http://localhost:5173 — paste code in the editor, click Scan Code.

Set VITE_SCANNER_URL to point to a remote server (defaults to http://localhost:3001):

VITE_SCANNER_URL=https://my-scanner.example.com npm run dev

VS Code Extension

The extension in vscode-extension/ scans on every file save and shows inline diagnostics.

Configuration (File → Preferences → Settings → AI Code Security Scanner):

| Setting | Default | Description | |---------|---------|-------------| | aiSecScan.serverUrl | http://localhost:3001 | Scanner server URL | | aiSecScan.apiKey | "" | Bearer token — required when SERVER_API_KEY is set on the server | | aiSecScan.autoScanOnSave | true | Scan active file on save |

Tests

npx ts-node tests/scanner.test.ts

Detectors

The scanner ships 32 finding types across 8 languages (TypeScript/JavaScript, Python, Go, Java, C/C++, C#, Kotlin, Ruby):

| # | Finding type | Severity | Languages | Description | |---|-------------|----------|-----------|-------------| | 1 | SECRET_HARDCODED | critical | all | API keys, tokens, and passwords assigned to variables | | 2 | SQL_INJECTION | critical | JS/TS, Python, Go, Java, Ruby | String concatenation or template literals inside SQL queries | | 3 | SQL_INJECTION_CS | critical | C# | SqlCommand built with string concatenation from user input | | 4 | XSS | critical | JS/TS | Unsanitised user input in innerHTML, dangerouslySetInnerHTML, or document.write | | 5 | SHELL_INJECTION | high | JS/TS | exec() / execSync() with template literals or concatenated user input | | 6 | EVAL_INJECTION | high | JS/TS, Python | eval() or new Function() with dynamic arguments | | 7 | PATH_TRAVERSAL | high | JS/TS, Python, Go | fs calls combined with path.join using unsanitised user input | | 8 | PATH_TRAVERSAL_CS | high | C# | File.* or Path.Combine with user-controlled path components | | 9 | PROTOTYPE_POLLUTION | high | JS/TS | Object.assign, _.merge, or bracket notation writes that can pollute __proto__ | | 10 | INSECURE_RANDOM | medium | JS/TS | Math.random() used in security-sensitive contexts (tokens, IDs, passwords) | | 11 | OPEN_REDIRECT | medium | JS/TS | res.redirect() with dynamic, unvalidated destination | | 12 | SSRF | high | JS/TS, Python, Go | fetch(), axios, or http.get() with dynamic, user-controlled URLs | | 13 | COMMAND_INJECTION | high | JS/TS | spawn() / spawnSync() with a dynamic, user-controlled command string | | 14 | COMMAND_INJECTION_C | critical | C/C++ | system() / popen() with user-controlled command string | | 15 | COMMAND_INJECTION_CS | critical | C# | Process.Start() / ProcessStartInfo with user-controlled command | | 16 | CORS_MISCONFIGURATION | high | JS/TS | Wildcard origin with credentials: true, or reflected req.headers.origin | | 17 | JWT_HARDCODED_SECRET | critical | JS/TS | jwt.sign() with a hardcoded string secret | | 18 | JWT_WEAK_SECRET | high | JS/TS | jwt.sign() with a short (< 32 char) secret | | 19 | JWT_NONE_ALGORITHM | high | JS/TS | jwt.verify() without an algorithms whitelist, or with algorithms: ['none'] | | 20 | JWT_DECODE_NO_VERIFY | high | JS/TS | jwt.decode() used instead of jwt.verify() — signature not checked | | 21 | REDOS | medium | JS/TS | new RegExp() constructed from dynamic (user-controlled) input | | 22 | WEAK_CRYPTO | medium | JS/TS, Python | crypto.createHash() using MD5, SHA-1, MD4, or other weak algorithms | | 23 | UNSAFE_DESERIALIZATION | critical | Python | pickle.loads() or equivalent with untrusted data | | 24 | INSECURE_ASSERT | medium | Python | Security check implemented with assert, which is stripped in optimised mode | | 25 | INSECURE_BINDING | medium | Python, Go | Server bound to 0.0.0.0, exposing the service on all interfaces | | 26 | XML_INJECTION | high | Python, Java | XML parser configured without disabling external entities (XXE) | | 27 | LDAP_INJECTION | high | Python, Java | LDAP query built with string concatenation from user-controlled input | | 28 | BUFFER_OVERFLOW | critical | C/C++ | Unsafe buffer operations (gets, strcpy, sprintf) without bounds checking | | 29 | FORMAT_STRING | critical | C/C++ | Non-literal format string passed to printf/fprintf family | | 30 | MASS_ASSIGNMENT | high | Ruby | permit(:all) or unrestricted parameter binding in Rails controllers | | 31 | SSTI | critical | Python | Template string rendered from user-controlled input (Jinja2, Mako) | | 32 | INSECURE_SHARED_PREFS | medium | Kotlin/Android | Sensitive data written to SharedPreferences without encryption | | 33 | WEBVIEW_LOAD_URL | high | Kotlin/Android | WebView.loadUrl() called with user-controlled input | | 34 | PERFORMANCE_N_PLUS_ONE | medium | Kotlin | ORM or DB query executed inside a loop — N+1 query pattern | | — | UNSAFE_DEPENDENCY | medium | JS/TS | package.json dependency pinned to *, latest, or x; or missing lockfile | | — | VULNERABLE_DEPENDENCY | critical/high/medium | JS/TS | Known-vulnerable package version (CVE checked against a built-in list) |

Use in CI

The scanner ships a reusable GitHub Actions workflow. Call it from any workflow:

jobs:
  security:
    uses: ./.github/workflows/security-scan.yml
    with:
      path: src/          # directory or file to scan (default: .)
      fail-on: high       # critical | high | any (default: high)

After the job completes:

SARIF report is uploaded to GitHub Security tab (Code Scanning) and as a downloadable artifact
JSON report is saved as a downloadable artifact
Outputs findings-count and critical-count for downstream steps

Inputs

| Input | Default | Description | |-------|---------|-------------| | path | . | File or directory to scan | | fail-on | high | Minimum severity that causes the job to fail |

Environment variables

| Variable | Description | |----------|-------------| | GITHUB_TOKEN | Used for GitHub Contents API when scanning repos (auto-provided by Actions) | | ANTHROPIC_API_KEY | Optional — enables AI explanations and fix suggestions via aiExplain flag | | SERVER_API_KEY | Required in production — Bearer token that callers must include in Authorization: Bearer <key>. If unset, server runs in open-access dev mode. | | VITE_SCANNER_URL | Web UI only — overrides the default scanner server URL (http://localhost:3001) |

Publish to npm

npm publish          # runs tsc automatically via prepublishOnly
npm publish --tag beta

The .npmignore excludes src/, web/, tests/, and *.map files — only dist/ is published.

Automated publishing via GitHub Actions is configured in .github/workflows/publish.yml — it triggers on GitHub releases and runs npm publish automatically.