securepulse
Continuous website security monitoring for agencies, freelancers and SMBs.
- Language
- TypeScript
- Stars
- 0
- Last updated
From the README
A site's security posture decays quietly. Certificates expire, security headers get dropped in a redeploy, DNS records drift during a migration. A one-off audit is out of date within two months. SecurePulse re-runs a twelve-point check every day and emails you the moment the grade falls.
Scope. SecurePulse performs passive configuration review only. It reads what a server already serves to any browser or crawler. It never exploits, fuzzes, brute-forces or port-scans anything. See Scanning policy.
Status
Every claim below was verified by running it. See docs/TEST-RESULTS.md for the raw output.
| Area | State |
|---|---|
| Scan engine (12 checks, weighted grade) | Working, 35 unit tests + live tests against real sites |
| Email + password auth, DB-backed sessions | Working, verified end to end in a browser |
| Password reset | Working, verified end to end. 7 tests |
| Domain management + DNS TXT verification | Working |
| Plan limits enforced server-side | Working, 14 tests |
| Stripe subscriptions | Working in test mode. No live keys in this build. 8 tests including forged-webhook rejection |
| Scheduled scans + email alerts | Implemented; alert logic and email persistence tested. Not yet exercised by a real cron in production |
| AI remediation summaries | Working, with a deterministic template fallback when no API key is set |
| Agency read-only API + key management | Working, verified end to end. 7 tests |
| CSV / JSON export, bulk scan | Working |
| Dashboard (overview, sites, reports, activity, settings) | Working, with command palette, dark mode and scan-to-scan diffs |
| Deployment | Configured for Vercel + Neon. Not deployed — no credentials were used |
Every feature listed on a plan in src/lib/billing/plans.ts is implemented
and enforced server-side. Nothing is sold that does not exist.
Not built: server-side PDF generation (the client report is a print-optimised page that saves to PDF from any browser) and email verification on signup. Both are in Remaining work.
Quick start
Requires Node 20+. Nothing else — no database to install, no credentials.
npm install npm run db:generate # only if you change the schema npm run db:migrate # builds the local database npm run dev # http://localhost:3000
Then open the site and scan a domain. It works immediately, with no .env at
all.
Optionally seed a development account:
npm run seed # demo@securepulse.local / correct horse battery stapleStop the dev server before running
seed,db:migrateordb:reset. The local database is PGlite, which is an embedded, single-process Postgres. A running dev server holds the data directory open, so writes made by a separate script are not visible to it until you restart. This only affects local development; withDATABASE_URLset there is no such constraint.
Configuration
Copy .env.example to .env.local. Every variable is documented there.
The app runs with nothing set. Each integration degrades to a clearly labelled disabled state rather than crashing:
| Unset | Behaviour |
|---|---|
DATABASE_URL |
Uses PGlite against .data/pg. Real Postgres, zero setup. |
STRIPE_* |
Billing disabled. The UI says so; it never implies payments work. |
RESEND_API_KEY |
Emails are written to the emails table with full bodies, and not transmitted. That table is the dev inbox. |
ANTHROPIC_API_KEY |
Remediation summaries use a deterministic template built from the same findings. |
In production, APP_SECRET and CRON_SECRET are required and the server
refuses to start without them.
What it checks
Twelve checks weighted to 100 points. The weighting reflects consequence, not effort.
| Category | Points | Checks |
|---|---|---|
| Encryption & certificates | 25 | HTTPS reachable and HTTP upgrade; certificate trusted and days to expiry; TLS version |
| Security headers | 34 | HSTS; CSP (graded on whether it would stop anything); clickjacking; nosniff; Referrer-Policy; Permissions-Policy |
| Cookies | 8 | Secure, HttpOnly, SameSite on cookies set by the homepage |
| Email & DNS | 15 | SPF (and whether it restricts); DMARC (and whether it enforces); CAA |
| Page content | 9 | Mixed content; outdated front-end libraries |
| Information disclosure | 6 | Exact software versions in response headers |
| Disclosure policy | 3 | security.txt |
| Exposed files | penalty | .env, .git, database dumps. Verified domains only. |
How the grade is computed
Three rules, each chosen so scores stay comparable across sites and over time, which is what makes the trend line and the regression alert mean anything:
- The denominator is fixed at 100. Every check contributes its points whether it passed, failed, or could not be assessed. If an unreachable site silently shrank the denominator, it could out-score a working one.
- Exposed files apply a flat penalty (30 points each) after the
weighted total, rather than costing a slice of one category. A publicly
readable
.envmust not leave a site looking healthy because its headers are tidy. - Grades: A+ ≥97, A ≥90, B ≥80, C ≥70, D ≥60, F below.
Real results from the live suite: github.com → B (87),
example.com → F (51), neverssl.com → F (25).
Architecture
src/
app/
page.tsx Landing page with the free public scanner
(auth)/ Sign in, sign up, forgot / reset password
dashboard/
page.tsx Overview: portfolio, issues, certificate watch
sites/ Site list, and detail with the trend chart
reports/ Every scan, paginated and filterable
scans/[id]/ Findings, evidence, diff against last scan
activity/ Audit log and the transactional email record
settings/ Profile, password, API keys, danger zone
billing/ Plan and Stripe portal
report/[id]/ Print-optimised client report
api/
auth/* Session auth and password reset
account/* Profile, password, API keys, deletion
domains/[id]/verify DNS TXT ownership proof
scans Authenticated scan (quota + rate limited)
scans/bulk Scan every site, stops cleanly at quota
scans/[id]/export CSV / JSON download
scan/public Anonymous scan (IP rate limited, passive only)
billing/{checkout,portal} Stripe
stripe/webhook The only thing that grants a paid plan
cron/scan Scheduled run, shared-secret authenticated
v1/* Agency read-only API, bearer-key authenticated
lib/
scanner/
hostname.ts Input parsing + SSRF address blocklist
fetcher.ts Budgeted HTTP, guarded DNS, TLS inspection
checks/ One module per category. Pure functions.
score.ts Weighting and grading
engine.ts Gathers evidence once, runs every check
auth/ scrypt hashing, opaque DB-backed sessions
billing/ Plan definitions, server-side limits, Stripe
email/ Transactional email, logged before dispatch
ai/remediation.ts Written summary, with template fallback
api-keys.ts Agency key issuance, auth and revocation
portfolio.ts Cross-site aggregates and scan-to-scan diffs
The scanner is pure. gatherEvidence() does all the network work once;
every check is a synchronous function from that evidence bundle to findings.
That is why the entire scoring system is testable against fixtures with no
network at all, and why adding a check cannot accidentally add a request.
Stack and why
| Layer | Choice | Reason |
|---|---|---|
| Next.js App Router | One deployable for marketing, app and API | |
| Drizzle + Postgres | SQL-first, no codegen step, same dialect everywhere | |
| PGlite in dev | Real Postgres in WASM. Zero-setup clone-and-run, and the same migrations run in production | |
| First-party auth | No third-party dependency or credential needed to run or test it. Opaque DB-backed sessions are revocable; a JWT is not | |
| scrypt | Ships in Node. No native module to compile, nothing in the auth path from npm | |
| undici | Lets the scanner validate resolved IPs at connect time, which is what closes the DNS-rebinding hole |
Security
Full review in docs/SECURITY.md. The parts specific to this product:
SSRF is the defining risk. Users hand us a hostname and the server then requests it. Defence is layered: IP literals and internal suffixes are rejected at parse time; a custom DNS lookup is wired into the connect path so the address that gets validated is the address the socket uses, which a resolve-then-fetch pre-flight cannot guarantee; and redirects are followed manually so every hop is re-validated. Verified by a live test against a public hostname that resolves to loopback — the connection is refused.
Abuse controls. Every scan is capped at 24 HTTP requests. Anonymous scans are rate-limited per IP and never run path discovery. Rate limiting is stored in Postgres with an atomic upsert, not an in-process map, because on serverless a per-instance counter is not a limit at all (tested: 10 concurrent requests against a limit of 5 yield exactly 5 allowed).
Billing integrity. The Stripe webhook is the only thing that grants a paid plan. Signatures are verified against the raw body, and entitlement is derived from the price that was billed, never from anything the request claims. Tested: an unsigned, forged, wrong-secret or tampered payload is rejected and the plan stays free.
Entitlement is evaluated on read. A past_due, canceled or lapsed
subscription resolves to free even if the stored plan column still says
pro, so a webhook we never received cannot leave someone on a paid plan
indefinitely.
Interface
The dashboard is built to be read quickly by someone who looks after a dozen sites they did not build:
- Overview rolls issues up by check rather than by site, because the answer to "what do I have to fix" is usually one header missing on nine of them. Certificates near expiry get their own list.
- Command palette (
Cmd-K/Ctrl-K) jumps to a site, runs a scan, or opens a page without navigating to find the control first.?lists every shortcut. - Reports show what changed since the previous scan, not only what is true now. That is the difference between monitoring and a spot check.
- Findings filter by status, severity and category, carry their evidence, and every fix is one click to the clipboard. Any scan exports to CSV or JSON.
- Dark mode is a selected palette, not an inversion: the accent is lifted and every status colour is re-stepped against the dark surface. It follows the system preference until you override it.
- Every status is carried by a word and a shape as well as a colour, so a report survives being printed in greyscale for a client.
Testing
npm test # 108 tests, no network, no credentials LIVE=1 npx vitest run tests/live # 11 more, against real sites npm run typecheck && npm run lint
Tests run against an in-memory PGlite database built by replaying the real
migration files, so a broken migration fails the suite instead of failing a
deploy. tests/setup.ts unsets DATABASE_URL, so the suite cannot reach a
real database.
Deployment
See docs/DEPLOYMENT.md for the full walkthrough (Vercel + Neon + Stripe + Resend, and the cron setup).
Nothing in this repository has been deployed, and no live credentials were used to build it.
Scanning policy
SecurePulse requests: the homepage over HTTPS and HTTP, a TLS handshake on
443, public DNS records, /.well-known/security.txt, and — only on
DNS-verified domains — a small fixed list of commonly-leaked paths.
It never exploits, fuzzes, brute-forces, port-scans, or sends volumetric traffic. Path discovery is gated on ownership proof specifically so the tool cannot be pointed at someone else's site as a reconnaissance service. Every scan is written to an audit log with its source address.
A good grade means the configuration a browser can see is sound. It is not a penetration test and says nothing about application code, dependencies, or who has access to the server. The product states this on the report itself.
Remaining work
Honest list of what a paying customer would notice missing:
- Server-side PDF. The client report is a print-optimised page. It works
and saves to PDF from any browser, but a one-click
.pdfdownload is what agencies will eventually ask for. - Email verification on signup. An account can be created with an address the user does not control. Low impact today, but it should be closed.
- Scheduled runs have not been exercised against a live cron. The logic is tested; the trigger has not run in production.
- Scan concurrency. The scheduled run processes domains serially with a batch cap of 25. Fine to a few hundred domains; needs a queue beyond that.
- Email deliverability. Requires a verified sending domain with SPF and DKIM before alerts will land reliably.
- Public report sharing was deliberately left out. A link that exposes a client's security weaknesses to anyone who has the URL needs expiry, revocation and a considered threat model. Half-building it would be worse than not having it.
Business documentation
- docs/DECISION.md — why this product, and why not the alternatives
- docs/GO-TO-MARKET.md — pricing rationale and customer acquisition plan
- docs/SECURITY.md — security review
- docs/TEST-RESULTS.md — what was actually run, and the output