web-frontend-security
Use when the diff renders user-controlled content as HTML, builds links or redirects, changes CSP or CORS, adds a cookie-authenticated state change or WebSocket, a postMessage handler, a markdown renderer, a service worker, browser storage of tokens or private data, a frameable sensitive action, or an Electron/Tauri window
- License 许可
- Apache-2.0
- 最近更新
- 2026-11-15
- Stars
- ★ 122
- Forks
- 18
01 前言 Front Matter
- name
- web-frontend-security
- source
- anthropics/claude-code-security-review (MIT) framework precedents, adapted; cloudflare/security-audit-skill (MIT) client-side attack classes (cross-site WebSockets, service workers, browser storage, DOM clobbering, UI redress), adapted; informed by anthropics/claude-code security-guidance plugin (ideas only); OWASP Top 10:2025 A05/A01 and CWE cited by name only
- category
- security
- tech_stack
- Web & API
- description
- Use when the diff renders user-controlled content as HTML, builds links or redirects, changes CSP or CORS, adds a cookie-authenticated state change or WebSocket, a postMessage handler, a markdown renderer, a service worker, browser storage of tokens or private data, a frameable sensitive action, or an Electron/Tauri window
03 SKILL.md 正文
Web Frontend Security
Overview
Modern template engines and UI frameworks escape by default. XSS comes back through the doors that switch escaping off, through the wrong escaper for the context, and through URLs. CSRF comes back whenever cookies authenticate a state change.
Core principle: find the escape hatch, then trace what flows into it. {value} in JSX is the control working — not a finding.
1. XSS Sinks by Framework (CWE-79)
| Framework / engine | Escape hatches to trace |
|---|---|
| React / Preact / Solid | dangerouslySetInnerHTML, innerHTML props, refs writing el.innerHTML |
| Vue | v-html, render functions setting innerHTML |
| Angular | bypassSecurityTrustHtml/Url/ResourceUrl/Script/Style ([innerHTML] alone is sanitised) |
| Svelte | {@html …} |
| DOM / jQuery | innerHTML, outerHTML, insertAdjacentHTML, document.write, .html(), $(userString), iframe.srcdoc |
| Jinja2 / Django | ` |
| Go | template.HTML(userValue), text/template used to render HTML |
| JVM | Thymeleaf th:utext, JSP <%= %> or escapeXml="false", FreeMarker ?no_esc |
| Node templates | Handlebars {{{ }}}, EJS <%- %>, Pug != |
| .NET | @Html.Raw, MarkupString |
// ❌ user-authored bio rendered as HTML
<div dangerouslySetInnerHTML={{ __html: user.bio }} />
// ✅ sanitised with an allowlist sanitiser, or rendered as text
<div dangerouslySetInnerHTML={{ __html: DOMPurify.sanitize(user.bio) }} />
Markdown renderers count as HTML sinks when raw HTML is enabled: marked without a sanitiser, markdown-it with html: true, react-markdown with rehype-raw and no rehype-sanitize.
URLs are sinks too. href, src, formaction, location.href = …, window.open(…) with user data accept javascript: (and data: for some sinks). Validate the scheme (https:, mailto:) rather than relying on the framework's version-dependent handling.
<!-- ❌ profile website link -->
<a :href="user.website">site</a>
<!-- ✅ -->
<a :href="safeHttpUrl(user.website)">site</a> <!-- returns "#" unless new URL(v).protocol is http: or https: -->
Wrong escaper for the context: HTML-escaping a value placed inside a <script> block, an event-handler attribute or a CSS value does not make it safe. Data embedded into an inline script needs JSON serialisation that escapes < (</script> breaks out).
Stored XSS (one user's content shown to another) → HIGH. Reflected → MEDIUM to HIGH by reach. Self-XSS → not a finding.
2. CSRF (CWE-352)
A finding when all hold: the endpoint changes state, the browser authenticates it automatically (cookies, HTTP auth), and nothing ties the request to the app's own pages (CSRF token, SameSite=Lax/Strict on the session cookie with no state-changing GET, an Origin/Sec-Fetch-Site check, a custom-header requirement the CORS policy enforces).
- APIs authenticated only by an
Authorization: Bearerheader are not CSRF-able. - A new state-changing
GETroute is CSRF-able even withSameSite=Lax. - Removing or bypassing the CSRF middleware for a route (
csrf_exempt,ignoringRequestMatchers) is a removed guard.
3. CORS
// ❌ any site can read authenticated responses: the request's Origin is reflected with credentials
res.setHeader("Access-Control-Allow-Origin", req.headers.origin);
res.setHeader("Access-Control-Allow-Credentials", "true");
// ✅ exact allowlist
if (ALLOWED_ORIGINS.has(req.headers.origin)) res.setHeader("Access-Control-Allow-Origin", req.headers.origin);
Access-Control-Allow-Origin: * without credentials is fine for public data; browsers refuse * with credentials. Watch for null origin allowed and suffix-matched origins (ssrf-and-url-allowlists has the matching pitfalls).
4. postMessage (CWE-346)
// ❌ any window can send a "token" message
window.addEventListener("message", (e) => { localStorage.setItem("token", e.data.token); });
// ✅
window.addEventListener("message", (e) => { if (e.origin !== "https://auth.example.com") return; … });
Also target.postMessage(secret, "*") — any framing or opening page can receive it.
5. CSP
A missing CSP is not reported. Adding 'unsafe-inline', 'unsafe-eval', a wildcard source or a JSONP-capable CDN to an existing policy is a weakened control: a note, or part of the finding when the same diff adds an injection sink.
6. Open Redirect (CWE-601) — high confidence only
Report only with a concrete chain: a post-login ?next= that can point off-site in an OAuth or SSO flow (code or token leakage), or a redirect used to pass an SSRF allowlist. A bare redirect on a marketing page is dropped.
7. Desktop Shells (Electron, Tauri)
In a desktop shell, renderer XSS can become code execution. This section covers the window and renderer settings; the main-process IPC handlers, URL schemes and loopback servers are desktop-and-local-ipc-security.
- Electron:
nodeIntegration: true,contextIsolation: false,sandbox: false,webSecurity: false; a preload exposingipcRendererorrequirewholesale throughcontextBridge;shell.openExternal(url)with an unvalidated scheme; IPC handlers that run commands or read paths from renderer arguments without validation; remote content loaded into a privileged window. - Tauri: broad
allowlist/capabilities (shellopen,fsscopes of**),dangerousRemoteDomainIpcAccess.
8. Cross-Site WebSocket Hijacking (CWE-1385)
A WebSocket upgrade carries the browser's cookies and no CORS check. A handshake authenticated by cookie that does not verify Origin against an exact allowlist (or require a token the page must fetch first) lets any site open the socket as the victim and read or send messages.
// ❌ cookie session, any origin
wss.on("connection", (ws, req) => { const user = sessionFrom(req.headers.cookie); … });
// ✅
const ok = ALLOWED_ORIGINS.has(req.headers.origin ?? "");
if (!ok) { ws.close(1008); return; }
Not a finding when the socket authenticates with a token sent in the first message or a header the page cannot be made to send.
9. Service Workers
- The script URL passed to
navigator.serviceWorker.registerbuilt from input, or served from a path user content can occupy; aService-Worker-Allowedheader that widens the scope beyond what the worker needs. - A fetch handler that caches personalized or authenticated responses without the user in the cache name or key, and serves them after logout or an account switch — or as the offline fallback for another user.
importScriptsof a third-party URL without integrity control.
10. Browser Storage and Account Switching
A token in localStorage is not a finding by itself (authn-sessions-and-tokens §3). It is when a realistic lower-authority reader exists, or when the data outlives the session: private responses, drafts or auth decisions left in localStorage, IndexedDB or Cache Storage after logout and shown to the next account on a shared machine; a storage event or BroadcastChannel message that carries a token or an auth command between tabs without binding it to the current session.
11. DOM Clobbering and Prototype-Pollution Gadgets
Both need two halves in the trace. DOM clobbering: markup the attacker controls that keeps an id or name attribute (a sanitizer that allows them) and code that reads the clobbered global as configuration (window.config || defaults, document.getElementById("cdn").href for a script URL). Prototype pollution: an attacker-chosen key reaching a recursive merge or path setter and a gadget that reads the polluted property in a security decision, a script URL or an HTML sink (node-typescript-security-pitfalls §5).
12. Clickjacking
A frameable page with a one-click state change — confirming a payment, granting an OAuth scope, deleting an account, changing an email — and no frame-ancestors or X-Frame-Options. Missing frame headers on read-only or multi-step pages are hardening, not reported.
Common Mistakes
- Flagging
{user.name}in JSX,{{ name }}in Vue/Jinja, or[innerHTML]in Angular. - Reporting CSRF on a bearer-token API.
- Reporting a missing CSP,
X-Frame-Optionsor HSTS as a vulnerability. - Reporting an open redirect with no chain.
Red Flags
- An escape hatch whose input traces back to another user's content.
DOMPurifyimported but the sink fed the unsanitised variable.Access-Control-Allow-Credentials: truenext to a reflected origin.addEventListener("message"with noorigincomparison.- An Electron
BrowserWindowcreated withnodeIntegration: trueor loading remote URLs. - A WebSocket server that reads the session cookie and never reads
Origin. - A service worker that
cache.puts responses from authenticated API routes. - A sanitizer configured to keep
id/namenext to code that readswindow.<name>.
References (names and links only)
OWASP Top 10:2025 A05 Injection, A01 Broken Access Control · ASVS 5.0 V3 Web Frontend Security · CWE-79, 346, 352, 601, 942, 1021, 1321, 1385 · Electron security checklist
04 同仓库更多 Skills
visual-craft
暂无简介
ux-copy
Use when you write any words a user reads in a mockup or a hand-off - content language and voice, buttons, labels, errors, empty states, confirmations, numbers and dates, and the longest-string check
screen-mockup-html
Use when you draw screens as the design task's HTML mockup - ONE canvas document holding every screen and every variant side by side, the sanitizer-safe recipe, phone and wide frames, the states grid, static control states, and the loopback self-check with screenshots at 375, 768 and 1440 attached to the task
performance-awareness
How the performance score moves for your role. Use when the score context shows a declining trend, or before handing off work you have not self-checked against every criterion.
design-variants
Use when a screen design needs variants - how many, variants as side-by-side rows on the one canvas, what makes two variants genuinely different, recommending one, how the human's choice is read, and narrowing to the chosen one in revision
design-system-authoring
Use when you derive, change or extend a design system - where tokens live per stack, the DTCG token tree, role-based names for both themes, base vs layer vs drift, the DESIGN.md section order, the inventory format, the contrast check and the lint every proposal returns