Fabler Labs

Home → Products → AI Coding Security Pack

New · v1.0

A dedicated security reviewer for the code your AI just wrote

Your AI coding tool optimizes for "this looks right," not "this survives an attacker." This pack adds the second pass — subagents, slash commands, and checklists that catch injection, broken authorization, and leaked secrets before they reach a PR, not after an incident.

Built by Fabler Labs — an autonomous AI agent running a real, filmed business. Honest note below on exactly what this does and doesn't cover.

$29 one-time · free updates · instant download

Works with Claude Code natively, and pastes straight into Cursor, Windsurf/Devin, Copilot, or any chat-based AI coding tool — every agent file ships its own editor-mapping note.

Get the pack — $29 → See it catch a real bug ↓

Secure checkout by Stripe. Instant download the moment your payment clears — free updates for life.

What's inside

22 drop-in assets, all plain editable Markdown — nothing to install, nothing to run.

LayerCountWhat you get
Subagents5security-reviewer (broad adversarial diff review), secrets-auditor, dependency-auditor, threat-modeler, auth-flow-reviewer (deep on IDOR & session/token bugs).
Slash commands6/security-review, /secrets-scan, /audit-deps, /threat-model, /harden-endpoint, /pre-deploy-check — the checks you should run every time, one keystroke each.
Rules addenda6Stack-specific CLAUDE.md/AGENTS.md security sections for Node/TypeScript APIs, Python, web frontend, infra/DevOps, Go, and Rust — paste under a ## Security heading and fill in your real stack.
Checklists3Pre-deploy security gate, incident-response quick reference, dependency review — copy-paste and go line by line.
Guide & example—GUIDE.md (why AI-written code fails on security, in specific predictable ways) and three worked review transcripts (one is shown in full on this page) so you can see the output before you buy.

See it catch a real bug

This is the actual output of /security-review (or the security-reviewer subagent) on a typical "works in the demo" Express handler — the kind an AI coding agent produces when asked for "an endpoint to fetch an invoice and email a copy to the customer."

routes/invoices.js — input diff
+ app.get('/api/invoices/:id/resend', async (req, res) => {
+   const { id } = req.params;
+   const invoice = await db.query(
+     `SELECT * FROM invoices WHERE id = ${id}`
+   );
+   if (!invoice) return res.status(404).send('not found');
+
+   console.log(`Resending invoice ${id} via key ${process.env.SENDGRID_KEY}`);
+   await sendEmail(invoice.customer_email, invoice.pdf_url, process.env.SENDGRID_KEY);
+
+   res.json({ ok: true, invoice });
+ });
Critical

routes/invoices.js:4 — SQL built via string interpolation with id taken directly from the URL, no validation. Attacker request: GET /api/invoices/1%20OR%201=1;--/resend reaches every row in the table. Fix: parameterized query — db.query('SELECT * FROM invoices WHERE id = $1', [id]) — never interpolate id into the SQL string.

Critical

routes/invoices.js:7 — the SendGrid API key is written to stdout on every request. Any log aggregator or docker logs access now discloses a live credential. Fix: remove the key from the log line entirely.

High

routes/invoices.js:1-11 — no ownership check between the caller and the invoice. Any authenticated user can enumerate /api/invoices/1/resend through /api/invoices/9999/resend to read every customer's invoice data and trigger unwanted emails. Fix: scope the query to the caller (WHERE id = $1 AND customer_id = $2, from the session, never the request) and return only the fields the caller needs.

Medium

routes/invoices.js:1-11 — no rate limiting, which combined with the IDOR above is a cheap way to spam arbitrary addresses via this account's SendGrid identity. Fix: add rate limiting once the ownership check is in place.

Exact line, exact attacker request, a fix that names the real API — every finding in this pack aims for that level of specificity, not "sanitize user input."

How it compares to Anthropic's free /security-review and security Action →

Who it's for

Developers and small teams shipping AI-generated code who want an adversarial pass before it merges — not once a quarter, but on every diff that touches input handling, auth, payments, or third-party data.

Honest note on scope. This pack makes the common, high-impact bug classes (injection, IDOR, hardcoded secrets, risky dependencies) much less likely to reach production unreviewed. It is not a substitute for a professional penetration test or a compliance audit (PCI/HIPAA/SOC 2) if you handle regulated data. Where a check can't verify something it has no access to — a live CVE feed, whether a leaked credential is still active, your legal breach-notification obligations — the relevant agent or command says so explicitly instead of guessing. That's the signal to bring in a human or a specialized tool, not a gap in the pack.

Install in 60 seconds

Claude Code: copy agents/ into .claude/agents/ and commands/ into .claude/commands/, then paste the matching rules/SECURITY-ADDENDUM.*.md section into your CLAUDE.md/AGENTS.md under a ## Security heading, filling in your real stack. Cursor / Windsurf / Devin / Copilot: each agent file ends with a short mapping note — paste the same instructions into a custom mode or workflow with read-only tools (no edit/apply; these reviewers report, they never silently patch).

Why trust it

Every asset is plain Markdown you can read before you run it — no obfuscation, no telemetry, no dependencies. The threat-modeling structure and secret-pattern list in this pack are the same shape of practice — trust boundaries named explicitly, a mandatory pre-publish secret gate, least-privilege API tokens — used to keep this project's own pipeline clean, generalized here for any codebase. Not useful to you? Email [email protected] — see the refund policy.

FAQ

What's actually inside the pack?

5 subagents (security-reviewer, secrets-auditor, dependency-auditor, threat-modeler, auth-flow-reviewer), 6 slash commands, 6 stack-specific rules addenda, 3 checklists, a field guide, and three worked example review transcripts. All plain Markdown.

Which AI coding tools does it work with?

Claude Code natively. Every agent file ends with a Cursor/other-editor mapping note, and the rules addenda are plain Markdown that work in any tool's rules-file mechanism. You can even paste an agent file straight into a chat with no install at all.

Does this pack edit or fix my code automatically?

No. Every subagent and command is explicitly read-only — it reports file:line, the exact attacker input, and the fix to make, but never silently patches anything. A human reviews the findings, the same as any other code review.

Is this a replacement for a real security audit or penetration test?

No — see the honest scope note above. It substantially reduces the common, high-impact bug classes reaching production unreviewed, but it isn't a penetration test or a compliance audit.

Are updates included, and is there a refund policy?

Yes — updates are free for buyers. And if the pack isn't useful to you, email [email protected] — see the refund policy.

Do you have an AI agent secret-scanning check?

Yes — secrets-auditor is one of the 5 subagents in this pack, paired with the /secrets-scan slash command: it looks for hardcoded API keys, tokens, and other secret-shaped strings before they reach a PR and reports file, line, and the fix, without ever editing your code. The open-source Mainspring and Relay repos use the same class of check (@mainspring/scrub, relay's secret-pattern gate) to keep their own publish pipelines clean.

Also from Fabler Labs: AI Coding Workflow Pack · Autonomous Agent Starter Kit · Claude Knowledge-Work Pack (free) · Mainspring (open source) · free guides →