Agent Skills: Authentication Flow Rules Skill

Authentication Flow Rules Skill

Iron Laws

1. PKCE IS MANDATORY FOR ALL CLIENTS — OAuth 2.1 requires PKCE for both public AND confidential clients; there are no exceptions and no legacy carve-outs. Every authorization code flow must generate and validate a code_challenge with method S256.
2. IMPLICIT FLOW IS PERMANENTLY REMOVED — Never use response_type=token; tokens returned in URL fragments leak via browser history, referrer headers, and server logs. Migrate immediately to Authorization Code + PKCE.
3. TOKENS MUST NEVER BE STORED IN LOCALSTORAGE OR SESSIONSTORAGE — XSS vulnerabilities can exfiltrate tokens from JavaScript-accessible storage. Store access tokens in HttpOnly, Secure, SameSite=Strict cookies only.
4. ACCESS TOKEN LIFETIME MUST NOT EXCEED 15 MINUTES — Short-lived tokens limit the blast radius of token theft. Refresh tokens must rotate on every use and invalidate all sessions on reuse detection.
5. EXACT REDIRECT URI MATCHING IS NON-NEGOTIABLE — No wildcards, no partial matches, no trailing slash tolerance. Authorization server must reject any redirect_uri that does not match the pre-registered value exactly.

Anti-Patterns

| Anti-Pattern | Why It Fails | Correct Approach | | ------------------------------------------------------------------------ | -------------------------------------------------------------------------------------------- | --------------------------------------------------------------------------------- | | Storing tokens in localStorage | Exposed to XSS; any script on the page (including third-party) can read and exfiltrate | Use HttpOnly cookies set server-side after token exchange | | Using Implicit Flow (response_type=token) | Tokens in URL fragments leak via browser history, Referer headers, and proxy/CDN logs | Use Authorization Code Flow with PKCE | | Collecting user passwords directly (Resource Owner Password Credentials) | Violates OAuth separation of concerns; client handles credentials it should never see | Use Authorization Code Flow; direct users to the authorization server login page | | Wildcard or partial redirect URI matching | Open redirect attack; adversary registers https://evil.com which prefix-matches a wildcard | Register exact URIs; server rejects any URI not in the pre-approved list | | Long-lived access tokens (>15 min) without rotation | Token theft window is unbounded; compromised token grants long-term access | Keep access tokens ≤15 min; implement refresh token rotation with reuse detection |

CRITICAL: Required Changes from OAuth 2.0

• PKCE is REQUIRED for ALL clients (public AND confidential)
• Implicit Flow is REMOVED - do not use, migrate immediately
• Resource Owner Password Credentials REMOVED - never collect user passwords directly
• Bearer tokens in URI query parameters FORBIDDEN - tokens only in Authorization headers or POST bodies
• Exact redirect URI matching REQUIRED - no wildcards, no partial matches

Authorization Code Flow with PKCE (The ONLY User Flow)

Step 1: Generate PKCE Challenge

// Client-side: Generate code_verifier (43-128 chars, URL-safe)
async function generatePKCE() {
  const array = new Uint8Array(32);
  crypto.getRandomValues(array);
  const verifier = base64UrlEncode(array);

  // Hash verifier to create challenge
  const encoder = new TextEncoder();
  const hash = await crypto.subtle.digest('SHA-256', encoder.encode(verifier));
  const challenge = base64UrlEncode(new Uint8Array(hash));

  return { verifier, challenge };
}

Step 2: Authorization Request

const { verifier, challenge } = await generatePKCE();
sessionStorage.setItem('pkce_verifier', verifier); // Temporary storage only

const authUrl = new URL(authorizationEndpoint);
authUrl.searchParams.set('response_type', 'code');
authUrl.searchParams.set('client_id', clientId);
authUrl.searchParams.set('redirect_uri', redirectUri); // MUST match exactly
authUrl.searchParams.set('code_challenge', challenge);
authUrl.searchParams.set('code_challenge_method', 'S256'); // SHA-256 required
authUrl.searchParams.set('scope', 'openid profile email');
authUrl.searchParams.set('state', cryptoRandomState); // CSRF protection

window.location.href = authUrl.toString();

Step 3: Token Exchange with Code Verifier

const response = await fetch(tokenEndpoint, {
  method: 'POST',
  headers: { 'Content-Type': 'application/x-www-form-urlencoded' },
  body: new URLSearchParams({
    grant_type: 'authorization_code',
    code: authorizationCode,
    code_verifier: sessionStorage.getItem('pkce_verifier'), // Prove possession
    client_id: clientId,
    redirect_uri: redirectUri, // MUST match authorization request
  }),
});

// Clear verifier immediately after use
sessionStorage.removeItem('pkce_verifier');

Token Security Best Practices

Token Lifetimes (RFC 8725)

• Access tokens: ≤15 minutes maximum
• Refresh tokens: Rotate on every use (sender-constrained or rotation required)
• ID tokens: Short-lived, validate signature and claims

Token Storage (CRITICAL)

// ✅ CORRECT: HttpOnly, Secure, SameSite cookies
// Server sets cookies after token exchange
res.cookie('access_token', accessToken, {
  httpOnly: true, // Prevents XSS access
  secure: true, // HTTPS only
  sameSite: 'strict', // CSRF protection
  maxAge: 15 * 60 * 1000, // 15 minutes
});

res.cookie('refresh_token', refreshToken, {
  httpOnly: true,
  secure: true,
  sameSite: 'strict',
  path: '/auth/refresh', // Limit scope
  maxAge: 7 * 24 * 60 * 60 * 1000, // 7 days, rotate frequently
});

// ❌ WRONG: NEVER store tokens in localStorage or sessionStorage
localStorage.setItem('token', accessToken); // VULNERABLE TO XSS
sessionStorage.setItem('token', accessToken); // VULNERABLE TO XSS

Refresh Token Rotation

// Server-side: Rotate refresh tokens on every use
async function refreshTokens(oldRefreshToken) {
  // Validate old token
  const userId = await validateRefreshToken(oldRefreshToken);

  // Detect token reuse (possible theft)
  if (await isTokenAlreadyUsed(oldRefreshToken)) {
    await revokeAllTokensForUser(userId); // Kill all sessions
    throw new Error('Token reuse detected - all sessions revoked');
  }

  // Mark old token as used BEFORE issuing new one
  await markTokenAsUsed(oldRefreshToken);

  // Issue new tokens
  const newAccessToken = generateAccessToken(userId, '15m');
  const newRefreshToken = generateRefreshToken(userId);

  return { newAccessToken, newRefreshToken };
}

REMOVED Flows (Do Not Use)

❌ Implicit Flow (REMOVED in OAuth 2.1)

// NEVER DO THIS - Implicit Flow is REMOVED
authUrl.searchParams.set('response_type', 'token'); // ❌ FORBIDDEN
// Tokens in URL fragments leak via browser history, referrer headers, logs

Migration Path: Use Authorization Code Flow + PKCE instead.

❌ Resource Owner Password Credentials (REMOVED)

// NEVER DO THIS - Collecting passwords directly violates OAuth
fetch(tokenEndpoint, {
  body: new URLSearchParams({
    grant_type: 'password', // ❌ FORBIDDEN
    username: user.email,
    password: user.password,
  }),
});

Migration Path: Use Authorization Code Flow for user auth, Client Credentials for service accounts.

Modern Authentication Patterns

Passkeys/WebAuthn (Recommended for 2026+)

// Register passkey
const credential = await navigator.credentials.create({
  publicKey: {
    challenge: serverChallenge,
    rp: { name: 'Your App' },
    user: {
      id: userIdBytes,
      name: user.email,
      displayName: user.name,
    },
    pubKeyCredParams: [{ alg: -7, type: 'public-key' }],
    authenticatorSelection: {
      authenticatorAttachment: 'platform', // Device-bound
      userVerification: 'required',
    },
  },
});

// Authenticate with passkey
const assertion = await navigator.credentials.get({
  publicKey: {
    challenge: serverChallenge,
    rpId: 'yourapp.com',
    userVerification: 'required',
  },
});

Security Checklist

Before deploying:

• [ ] PKCE enabled for ALL clients (no exceptions)
• [ ] Implicit Flow completely removed/disabled
• [ ] Password Credentials Flow removed/disabled
• [ ] Exact redirect URI matching enforced (no wildcards)
• [ ] Tokens NEVER in URL query parameters
• [ ] Access tokens ≤15 minutes
• [ ] Refresh token rotation enabled with reuse detection
• [ ] Tokens stored in HttpOnly, Secure, SameSite=Strict cookies
• [ ] All token transport over HTTPS only
• [ ] State parameter used for CSRF protection
• [ ] Token signature validation (JWT RS256 or ES256, not HS256)

Common Implementation Patterns

Login with OAuth 2.1 (email/password)

// Use Authorization Code Flow with PKCE even for first-party login
async function login(email: string, password: string) {
  // 1. Start PKCE flow
  const { verifier, challenge } = await generatePKCE();
  sessionStorage.setItem('pkce_verifier', verifier);

  // 2. POST credentials to your auth server's login endpoint
  const response = await fetch('/auth/login', {
    method: 'POST',
    headers: { 'Content-Type': 'application/json' },
    body: JSON.stringify({ email, password, code_challenge: challenge }),
  });

  // 3. Server validates credentials, returns authorization code
  const { code } = await response.json();

  // 4. Exchange code for tokens with verifier
  await exchangeCodeForTokens(code);
}

Login with GitHub OAuth

async function loginWithGitHub() {
  const { verifier, challenge } = await generatePKCE();
  sessionStorage.setItem('pkce_verifier', verifier);

  const authUrl = new URL('https://github.com/login/oauth/authorize');
  authUrl.searchParams.set('client_id', GITHUB_CLIENT_ID);
  authUrl.searchParams.set('redirect_uri', 'https://yourapp.com/auth/callback');
  authUrl.searchParams.set('scope', 'user:email');
  authUrl.searchParams.set('state', generateState());
  // Note: GitHub doesn't support PKCE yet, but use it when available

  window.location.href = authUrl.toString();
}

Logout

async function logout() {
  // 1. Clear client-side cookies
  document.cookie = 'access_token=; Max-Age=0; path=/';
  document.cookie = 'refresh_token=; Max-Age=0; path=/';

  // 2. Revoke tokens on server
  await fetch('/auth/logout', { method: 'POST' });

  // 3. Redirect to login
  window.location.href = '/login';
}

Memory Protocol (MANDATORY)

Before starting:

cat .claude/context/memory/learnings.md

After completing: Record any new patterns or exceptions discovered.

ASSUME INTERRUPTION: Your context may reset. If it's not in memory, it didn't happen.