JWT Attack Detection

SkillSecurity

Detect JWT implementation vulnerabilities including algorithm confusion, none algorithm acceptance, weak secrets, and JWK injection attacks.

Available today. Use it from your connected AI after setup.

Connect ahel once, and every AI you use reads what you have installed.

Then ask your AI: use the JWT Attack Detection skill

What this skill tells your AI

The instructions your AI receives, as published by byamb4/find-cve-agent in skills/jwt-attacks/SKILL.md and read by ahel’s review.

When to Use

Audit JWT verification/generation libraries, authentication implementations, and any code that validates or creates JSON Web Tokens.

Attack Types

1. Algorithm Confusion (RS256 to HS256)

The server uses RS256 (asymmetric) but the attacker changes the token header to HS256 (symmetric) and signs with the public key as the HMAC secret.

Conditions: Library accepts algorithm from token header without allowlist validation.

2. alg:none Bypass

Token header specifies "alg": "none", and the library accepts unsigned tokens.

Conditions: Library does not validate algorithm or allows "none".

3. JWK Header Injection

Attacker embeds their own public key in the token header via the jwk parameter, and the library uses it for verification.

4. Weak HMAC Secrets

HMAC secrets that are short, common words, or default values. Can be brute-forced offline.

5. kid (Key ID) Attacks

  • Path traversal: "kid": "../../dev/null" -- sign with empty key
  • SQL injection: "kid": "' UNION SELECT 'secret' --" -- inject known key
  • Command injection: "kid": "|id" -- if kid is passed to shell

6. jku/x5u URL Manipulation

jku (JWK Set URL) or x5u (X.509 URL) in header points to attacker-controlled server hosting a JWK Set with the attacker key.

Process

Step 1: Find JWT Usage

grep -rn "jwt\.verify\|jwt\.decode\|jwt\.sign\|jwt\.encode" .
grep -rn "jsonwebtoken\|jose\|PyJWT\|go-jose\|nimbus-jose" .
grep -rn "JWTVerify\|jwtVerify\|createRemoteJWKSet" .

Step 2: Check Algorithm Validation

grep -rn "algorithms\|algorithm.*=\|alg.*:" . | grep -i jwt

Is the algorithm explicitly specified or taken from the token header?

// VULNERABLE: no algorithm specified
jwt.verify(token, key);

// SAFE: algorithm allowlist
jwt.verify(token, key, { algorithms: ['RS256'] });

Step 3: Check for none Algorithm

grep -rn "none\|None\|NONE" . | grep -i "alg\|algorithm"

Step 4: Check kid Parameter Handling

grep -rn "kid\|keyId\|key_id\|getKey\|keyStore" .

Is the kid value used in file paths, database queries, or command execution?

Step 5: Check Secret Strength

grep -rn "secret\|SECRET\|JWT_SECRET\|TOKEN_SECRET" .

Is the secret hardcoded, from environment variable, or sufficiently random?

CVSS Guidance

  • Algorithm confusion to forge admin tokens: CRITICAL 9.1
  • alg:none bypass: CRITICAL 9.1
  • JWK injection: HIGH 8.1
  • Weak HMAC secret (brute-forceable): HIGH 7.5
  • kid path traversal: HIGH 7.5-8.1
  • kid SQLi: CRITICAL 9.8

References

Signals

GitHub stars
50
Forks
9
Last commit
Mar 2026
Advanced
Catalog kind
skill
Gateway key
jwt-attacks-byamb4
Source
github.com/byamb4/find-cve-agent