Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    Zero-Trust Frontend Security Audit

    1

    Adversarial frontend security audit: 202-item catalog, auditor → refuter → judge, zero-trust backend.

    Free

    1 useSecurity scanned
    Zero-Trust Frontend Security Audit

    Zero-Trust Frontend Security Audit

    Example session with this skill installed

    Do a full security audit of this Next.js frontend. Start with the deterministic recon so I can see the raw candidates.

    • Read your context and instructions
    • Compiled the zero-trust frontend security
    • Generated the document

    Summary

    Recon and candidate scan finished on the react-next test fixture that ships with the skill (18 files). The scanner produced

    24 candidates and the dependency checker

    8 findings, including a

    malicious axios release. These are candidates, not findings yet: next, each one goes through the auditors, the independent refuter and the judge.

    Dependencies (check-deps.mjs)

    SeverityPackageIssue
    criticalaxios@1.14.1Malicious publish (MAL-2026-2307) pulling a postinstall RAT. Remove, reinstall from a clean lockfile, rotate credentials
    criticalnext@15.2.0CVE-2025-66478, React2Shell (RSC RCE)
    criticalnext@15.2.0CVE-2025-29927, middleware bypass via x-middleware-subrequest
    highnext@15.2.0CVE-2025-55184, RSC DoS / source exposure
    mediumdompurify@3.3.1CVE-2026-0540, rawtext elements sanitization bypass
    medium(hygiene)No lockfile; .npmrc without ignore-scripts=true

    Top code candidates (scan.mjs)

    IDSevCatalogLocationPattern
    C-0001highFE-DATA-001.env:2Secret-looking variable with NEXT_PUBLIC_ prefix
    C-0024highFE-SSR-006app/profile/page.tsx:7Server-side fetch of a request-controlled URL
    C-0017highFE-MSG-001components/Bridge.tsx:7message listener without exact origin check
    C-0016highFE-XSS-001components/Bridge.tsx:9innerHTML = e.data.html
    C-0019highFE-FW-001components/NotificationList.tsx:13dangerouslySetInnerHTML with API data
    C-0015mediumFE-URL-002lib/redirect.ts:4Open redirect from ?next=
    C-0011mediumFE-AUTH-001lib/auth.ts:5Access token in localStorage
    C-0010mediumFE-ZT-010lib/auth.ts:10Decoded JWT claims used for authorization
    C-0003mediumFE-HDR-002next.config.js:14CSP with 'unsafe-inline' 'unsafe-eval'

    …plus 15 more medium candidates (unvalidated API casts, CSPT in API paths, Math.random nonce, wildcard next/image host, auth only in middleware).

    Next steps

    1. fe-recon builds the threat model, then 6 fe-auditor agents sweep the categories in parallel and promote candidates only with a cited source → sink trace.
    2. fe-refuter challenges every claim; disputes go to fe-judge.
    3. You get report.md + findings.sarif in .frontend-security-audit/, then approve the IDs you want fixed and fe-fixer applies and re-verifies them.

    page.tsx

    PDF · document

    Generated

    Example file from a real run - the skill writes it into your workspace.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Generic AI security reviews treat the frontend as an afterthought: they assume "React escapes everything", downplay open redirects and prototype pollution, and accept "the backend sanitizes it" as a defense. This skill runs an adversarial, evidence-based audit of your frontend codebase and assumes your backend is already compromised.

    Built by Antonio Luis Rossi, author of Segurança Front-end (Casa do Código). Open source (MIT): https://github.com/luisrossi/frontend-security-agent-skills

    What it does

    • 202-entry vulnerability catalog in 21 categories (DOM XSS, framework escape hatches, URL/navigation, postMessage, prototype pollution, CSP/Trusted Types, CORS/CSRF/XS-Leaks, auth/session, OAuth/OIDC in SPAs, sensitive data exposure, SSR/RSC, supply chain, LLM output rendering and more), each mapped to CWE, OWASP Top 10 2025 and ASVS 5.0 with detection, validation, false-positive conditions and fix code.
    • Auditor → refuter → judge pipeline. Parallel category auditors must cite a source → sink trace (file:line). An independent refuter, who never sees the auditor's reasoning, tries to kill each finding with concrete evidence. Disputes go to a one-round judge.
    • Dependency risk triage. Combines Dependabot alerts, npm audit, OSV and a curated list of malicious releases, then checks whether the vulnerable functions are actually reachable in your app. Returns a P0–P3 upgrade plan.
    • Deterministic tooling (Node 18+, zero dependencies): stack detector, candidate scanner (147 rules), dependency checker, security header evaluator, SARIF 2.1.0 exporter, Markdown report and a loopback-only Playwright PoC runner.
    • Fixes only with your approval, then re-verified.

    Frameworks

    React/Next.js, Vue/Nuxt, Angular, Svelte/SvelteKit, Astro, vanilla JS/jQuery/Lit/htmx, plus playbooks for 15 high-risk libraries (axios, lodash, DOMPurify, marked, markdown-it, react-markdown, qs, jwt-decode, js-cookie, socket.io, pdf.js and others).

    Safety

    The audit writes only to .frontend-security-audit/ in your project. PoCs use benign canaries and run only against loopback URLs you provide. Nothing is sent to third-party or production systems.

    Known limitations

    Requires Node.js 18+. Not for backend-only code, mobile native apps, IaC or smart contracts. Subagent parallelism needs an agent with subagents (e.g. Claude Code); other agents run the same roles sequentially.

    Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Frontend & Web Apps.

    Requires Node.js 18+. Audits are restricted to frontend code; backend, native mobile, and smart contracts are out of scope. Dynamic PoCs require a local URL.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before buying a skill.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    1 use

    Times used to date

    Free to use

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 25 days ago

    • Free with an Agensi account

    Accounts
    GitHubOsv
    Software
    Node.js 18+@lavamoat/allow-scriptsretirelockfile-lintzizmor@cyclonedx/cyclonedx-npmgitleaksfix-react2shell-nextGit
    Optional
    Playwrightosv-scannerGitHub CLI

    Permissions

    Terminal / Shell
    Read Files
    Browser
    Network Access
    Write Files
    Environment Variables

    Allowed Hosts

    github.com
    osv.dev

    Shell: runs the bundled Node.js scripts (scanner, dependency/header checks, report). Read files: audits your source code. Write files: writes only to .frontend-security-audit/; project files change only for fixes you approve. Network: optional OSV / GitHub Dependabot lookups and header checks. Browser: optional Playwright PoCs, loopback URLs only.

    Listed25 days ago

    What's inside

    Frequently Asked Questions

    Generic prompts tend to accept "the backend sanitizes it" and flag patterns without proof. This skill assumes the backend is compromised, requires a cited source → sink trace (file:line) for every finding, and sends each one to an independent refuter (and a judge on disputes), so false positives are filtered out before they reach your report.

    Node.js 18+ is required to run the bundled scripts. Playwright, osv-scanner, npm and the GitHub CLI (for Dependabot alerts) are optional. It follows the open SKILL.md standard: Claude Code runs the auditors in parallel as subagents; Codex CLI, Cursor, Gemini CLI, Copilot and other agents run the same roles sequentially.

    Both. Static findings must be proven with a code trace and survive an independent refuter. If you give it a running local URL and opt in, it can also confirm findings dynamically with Playwright, using benign canaries and only against loopback hosts. It never touches production or third-party systems.

    Yes, it's free and open source (MIT). You get the full skill: the 202-entry catalog, framework and library playbooks, the five agents (recon, auditor, refuter, judge, fixer), the zero-dependency Node.js scripts, and Markdown + SARIF reporting. Source: github.com/luisrossi/frontend-security-agent-skills

    The skill is designed to handle modern SPAs and Meta-frameworks including React, Next.js, Vue, Nuxt, Angular, Svelte, and Astro, covering both client-side logic and SSR-specific security concerns.

    New in Frontend & Web Apps

    Free