prodlint

Production readiness for vibe-coded apps.
Static analysis for vibe-coded apps. Flags the security, reliability, performance, and AI quality issues common in code from tools like Claude Code, Cursor, v0, Bolt, and Copilot: hallucinated imports, missing auth, hardcoded secrets, unvalidated server actions, and more. Zero config, no LLM, 52 rules.
prodlint v0.10.8
Scanned 6 files · 3 critical · 2 warnings · 1 info
app/api/users/route.ts
3:1 CRIT API route has no authentication check auth-checks
↳ Add authentication check: const session = await auth(); if (!session) return new Response('Unauthorized', { status: 401 })
4:1 WARN Request body accessed without validation (consider using Zod, Yup, or similar) input-validation
↳ Validate input with Zod or a similar schema library before using it
3:1 INFO No rate limiting — anyone could spam this endpoint and run up your API costs rate-limiting
↳ Add rate limiting: import { Ratelimit } from '@upstash/ratelimit' or use express-rate-limit
3:1 WARN API route handler has no try/catch block error-handling
↳ Wrap the handler body in try/catch and return appropriate error responses
lib/format.ts
1:1 CRIT Package "currency-format-utils" is imported but not in package.json hallucinated-imports
↳ Verify the package exists on npm. The AI may have invented this package name.
lib/stripe.ts
3:35 CRIT Hardcoded Stripe secret key detected secrets
↳ Move the secret to an environment variable and access via process.env.SECRET_NAME
Scores
security 82 ████████████████░░░░ (4 issues)
reliability 90 ██████████████████░░ (2 issues)
performance 100 ████████████████████ (0 issues)
ai-quality 100 ████████████████████ (0 issues)
Overall: 90/100
3 critical · 2 warnings · 1 info
Add to your README:
[](https://prodlint.com)
That's what prodlint prints for the six-file app in tests/fixtures/readme-demo. A test fails if this block and the scanner ever disagree.
Why?
Vibe coding is the fastest way to build. Shipping fast means knowing your code is production-ready — not just that it compiles. Hardcoded secrets, hallucinated packages, missing auth, and XSS vectors pass type-checks and look correct — but they aren't ready for production.
prodlint checks what TypeScript and ESLint don't: whether your vibe-coded app is ready for production. It runs next to them, not instead of them.
Install
npx prodlint
npx prodlint ./my-app
npx prodlint --json
npx prodlint --sarif
npx prodlint --summary
npx prodlint --profile startup
npx prodlint --profile balanced
npx prodlint --profile strict
npx prodlint --min-severity warning
npx prodlint --baseline-save .prodlint-baseline.json
npx prodlint --baseline .prodlint-baseline.json
npx prodlint --ignore "*.test.ts"
npx prodlint --quiet
npx prodlint --web example.com
Or install it:
npm i -D prodlint
npm i -g prodlint
Node.js 18 or later.
Exit codes
| Code | Meaning |
|---|
0 | No critical findings among the ones shown |
1 | At least one critical finding |
2 | The scan couldn't run: a bad flag, file or URL, or an unexpected error |
--min-severity, --profile and --baseline decide which findings are shown, and the exit code follows what's shown. Scores are always worked out from every finding.
JSON output
--json prints one object:
{
"version": "0.10.8",
"scannedPath": ".",
"filesScanned": 6,
"scanDurationMs": 39,
"findings": [
{
"ruleId": "secrets",
"file": "lib/stripe.ts",
"line": 3,
"column": 35,
"message": "Hardcoded Stripe secret key detected",
"severity": "critical",
"category": "security",
"fix": "Move the secret to an environment variable and access via process.env.SECRET_NAME"
}
],
"overallScore": 90,
"categoryScores": [
{ "category": "security", "score": 82, "findingCount": 4 }
],
"summary": { "critical": 3, "warning": 2, "info": 1 }
}
fix is there on every finding today, but treat it as optional.
52 Rules across 4 Categories
Each rule has a page with an example and a fix at https://prodlint.com/rules/<rule-id>.
Security (27 rules)
| Rule | What it checks |
|---|
secrets | Hardcoded API keys, tokens, private keys and database URLs with credentials |
auth-checks | API routes with no authentication check (middleware-aware) |
env-exposure | Server env vars read in client components; .env not in .gitignore |
input-validation | Request body used without validation in an API route |
cors-config | Access-Control-Allow-Origin: *, cors() with no config, origin: true; critical when the wildcard comes with credentials |
unsafe-html | dangerouslySetInnerHTML, direct innerHTML assignment |
sql-injection | SQL built with template literals or string concatenation (ORM-aware) |
open-redirect | User input passed directly to redirect() |
rate-limiting | API routes with no rate limiter |
phantom-dependency | Dependencies with commonly hallucinated or suspicious package names (slopsquatting) |
insecure-cookie | Session cookies set without httpOnly, secure or sameSite |
leaked-env-in-logs | process.env.* inside console.log() calls |
insecure-random | Math.random() used for tokens, secrets, or session IDs |
next-server-action-validation | Server actions using formData without Zod/schema validation |
env-fallback-secret | Secret env vars with hardcoded fallback values |
verbose-error-response | Error stack traces and messages sent back in API responses |
missing-webhook-verification | Webhook routes without signature verification |
server-action-auth | Server actions that write data, send something out or call an AI model, with no auth check |
eval-injection | eval(), new Function(), and strings passed to setTimeout or setInterval |
next-public-sensitive | NEXT_PUBLIC_ prefix on secret env vars |
ssrf-risk | User-controlled URLs passed to fetch in server code |
path-traversal | File system operations with unsanitized user input |
unsafe-file-upload | File uploads without type or size validation |
supabase-missing-rls | CREATE TABLE in migrations without enabling RLS |
deprecated-oauth-flow | OAuth Implicit Grant (response_type=token) |
jwt-no-expiry | JWT tokens signed without an expiration |
client-side-auth-only | Password comparisons, or auth tokens kept in localStorage, in client components |
Reliability (11 rules)
| Rule | What it checks |
|---|
hallucinated-imports | Imports of packages that aren't in package.json |
error-handling | API route handlers with no try/catch |
unhandled-promise | fetch and database calls with no await, return or assignment |
shallow-catch | Catch blocks that are empty, ignore the error, or log without it |
missing-loading-state | Client components that fetch without a loading state |
missing-error-boundary | Route layouts with no error.tsx beside them |
missing-transaction | Multiple Prisma writes without $transaction |
redirect-in-try-catch | redirect() inside try/catch — Next.js redirect throws, catch swallows it |
missing-revalidation | Server actions that change data with no revalidatePath or revalidateTag |
missing-useeffect-cleanup | useEffect with timers, listeners or subscriptions and no cleanup |
hydration-mismatch | window/Date.now()/Math.random() in server component render path |
| Rule | What it checks |
|---|
no-sync-fs | readFileSync and other synchronous fs calls in server code |
no-n-plus-one | Database or fetch calls inside loops |
no-unbounded-query | .findMany() / .select('*') with no limit |
no-dynamic-import-loop | import() inside loops |
server-component-fetch-self | Server components fetching their own API routes |
missing-abort-controller | fetch or axios calls with no timeout, in API routes, server actions and background jobs |
AI Quality (8 rules)
| Rule | What it checks |
|---|
ai-smells | TODO and FIXME comments, piles of console.log and any, commented-out code, "not implemented" stubs |
placeholder-content | Lorem ipsum, example emails, "John Doe" and "your-api-key-here" left in shipped code |
hallucinated-api | .flatten(), .contains(), .substr() and other methods that don't exist or are deprecated |
stale-fallback | Hardcoded localhost URLs, including as the fallback for an env var |
comprehension-debt | Functions over 80 lines, nested more than 5 levels deep, or with more than 5 parameters |
codebase-consistency | Conflicting conventions across files: export naming, import style, HTTP client, async style, quotes |
dead-exports | Exported functions, constants and types that nothing imports |
use-client-overuse | "use client" on files that don't use any client-side APIs |
Smart Detection
prodlint avoids common false positives:
- AST parsing — Babel-based analysis for 15 rules (imports, catch blocks, redirects, redirects in try/catch, SSRF, path traversal, JWT, HTML injection, hydration, transactions, env leaks, loops, SQL, route handlers, server actions) with regex fallback
- Monorepo support — npm/yarn/pnpm workspace dependencies resolved automatically
- Framework awareness — Prisma, Drizzle, Supabase, Knex, and Sequelize whitelists prevent false SQL injection flags
- Middleware detection — Clerk, NextAuth, Supabase middleware (
middleware.ts, or proxy.ts on Next.js 16) detected, matcher included — auth findings it covers drop to info
- Block comment awareness — patterns inside
/* */ are ignored
- Path alias support —
@/, ~/, and tsconfig paths aren't flagged as hallucinated imports
- Route exemptions — auth, webhook, health, and cron routes are skipped where a check wouldn't make sense
- Test and script awareness — most rules skip test files, and a missing package in a test or script is a warning, not critical
- Fix suggestions — every finding comes with a
fix hint
Scoring
Each category starts at 100. Deductions per finding:
| Severity | Deduction | Per-rule cap |
|---|
| critical | -8 | max 1 |
| warning | -2 | max 2 |
| info | -0.5 | max 3 |
Diminishing returns: after 30 points deducted in a category, further deductions are halved; after 50, quartered.
Weighted overall: security 40%, reliability 30%, performance 15%, ai-quality 15%. Floor at 0.
The caps mean one noisy rule can't sink a score, and also that a score can stay high with critical findings in it: the sample at the top is 90 with three. Read the findings, and treat the exit code as the gate.
GitHub Action
Add to .github/workflows/prodlint.yml:
name: Prodlint
on: [pull_request]
permissions:
contents: read
pull-requests: write
jobs:
scan:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v7
- uses: prodlint/prodlint@v1
with:
threshold: 50
Posts a score breakdown as a PR comment, and fails the build when the scan finds critical
issues or the score is below threshold. Each release of the action runs its matching
prodlint version, so pinning the action (@v1, a version tag, or a commit SHA) also pins
the scanner.
| Input | Default | Description |
|---|
path | . | Path to scan |
threshold | 0 | Minimum score to pass (0-100) |
fail-on-critical | true | Fail when the scan finds critical issues; false reports them without failing |
ignore | | Comma-separated glob patterns to ignore |
comment | true | Post PR comment with results |
| Output | Description |
|---|
score | Overall score (0-100) |
critical | Number of critical findings |
SARIF + GitHub Code Scanning
Upload prodlint results to GitHub's Security tab. The job needs security-events: write, and || true keeps a critical finding from stopping the job before the upload:
- name: Run prodlint
run: npx prodlint --sarif > prodlint.sarif || true
- name: Upload SARIF
uses: github/codeql-action/upload-sarif@v4
with:
sarif_file: prodlint.sarif
category: prodlint
Baseline for Existing Projects
Adopt prodlint gradually without drowning in pre-existing findings:
npx prodlint --baseline-save .prodlint-baseline.json
npx prodlint --baseline .prodlint-baseline.json
MCP Server
Use prodlint inside Cursor, Claude Code, or any MCP-compatible editor:
Claude Code:
claude mcp add prodlint -- npx -y prodlint-mcp
Cursor / Windsurf / Claude Desktop:
{
"mcpServers": {
"prodlint": {
"command": "npx",
"args": ["-y", "prodlint-mcp"]
}
}
}
VS Code (.vscode/mcp.json; the key is servers):
{
"servers": {
"prodlint": {
"command": "npx",
"args": ["-y", "prodlint-mcp"]
}
}
}
prodlint-mcp is a thin launcher for the server built into prodlint. It's published with
every prodlint release, so it always runs the matching scanner.
Ask your AI: "Run prodlint on this project" and it calls the scan tool directly.
| Tool | Inputs | Returns |
|---|
scan | path (project directory), optional ignore, minSeverity (critical / warning / info), categories, maxFindings (default 30) | Score, category scores, and findings most severe first. Every critical finding is listed, even past maxFindings. |
scan-web | url | Site score, grade, and the 14 agent-readiness checks |
Both tools return a Markdown summary plus structured output (structuredContent matching
each tool's outputSchema), and both are marked read-only (readOnlyHint), which hosted
reviewers such as GitHub Copilot code review require before they call an MCP tool.
scan reads files and sends nothing anywhere. It only scans a path inside the directory the
server was started in, so start it from your project. scan-web fetches the site you name,
from your machine.
Site Score
Check any deployed website for AI agent-readiness — 14 checks covering emerging standards like llms.txt, TDMRep, AgentCard, AI-Disclosure, HTTP Signatures (RFC 9421), and more.
npx prodlint --web example.com
npx prodlint --web example.com --json
prodlint site score
example.com · 14 checks
Score: 8 F ██░░░░░░░░░░░░░░░░░░
✗ robots.txt 0/7 No robots.txt found.
✗ AI robots.txt Directives 0/15 No robots.txt found. AI bots have no guidance.
✗ Content-Usage (IETF aipref) 0/5 No Content-Usage directives found.
✗ llms.txt 0/6 No llms.txt found.
✗ ai.txt 0/3 No ai.txt found at site root.
✗ TDMRep (W3C) 0/4 No TDMRep configuration found.
✗ AI-Disclosure Header 0/2 No AI-Disclosure header found.
✗ A2A AgentCard 0/5 No A2A AgentCard found.
✗ WebMCP Tools 0/5 No WebMCP tools detected.
✗ HTTP Message Signatures (RFC 9421) 0/3 No HTTP signature support detected.
✗ Structured Data 0/12 No structured data found.
✗ OpenGraph & Meta 0/10 No OpenGraph tags or meta description.
✗ sitemap.xml 0/10 No sitemap.xml found.
✓ Page Load Time 8/8 0.1s.
1 passed · 13 failed
Full results: https://prodlint.com/score?url=example.com
That's a real run: example.com is one bare page, so the only check it passes is answering quickly. Six of the checks are for settled things (robots.txt, rules for AI crawlers, structured data, Open Graph tags, a sitemap, response time) and carry 62 of the 95 points. The other eight are for drafts and proposals that may still change. A partial result earns half a check's points, rounded down, and the total is scaled to 100.
Or check your score interactively at prodlint.com/score.
prodlint is built for the patterns AI coding tools tend to leave behind. Every rule checks for something that breaks in production, not a style nit.
Suppression
Suppress a single line:
const key = "sk_test_example_for_docs"
Suppress an entire file (place at top):
Programmatic API
import { scan } from 'prodlint'
const result = await scan({ path: './my-project', ignore: ['legacy/**'] })
console.log(result.overallScore)
console.log(result.findings)
The package is ESM. It also exports rules, runWebScan, and the reporters reportPretty, reportJson, reportSummary and reportSarif.
Badge
[](https://prodlint.com)
License
MIT