Elegant JWT

JSON Web Tokens in the Elegant Objects style. The library hides pyjwt behind small immutable objects: a Token, its Claims, and a Signature that owns the algorithm and the key.

Table of Contents

• Installation
• Quick Start
• Tokens That Expire
• Refusing Expired Tokens
• Stamping the Issuer
• Tokens Valid Only Later
• Tokens for One Audience
• Asymmetric Algorithms
• Testing Without Waiting
• Your Own Signature
• Errors
• Design
• Development
• How to Report Issues

Installation

uv add elegant-jwt

or

pip install elegant-jwt

Quick Start

Create a token and read it back:

from elegant_jwt import Hs256, JwtClaims, JwtToken

signature = Hs256("a-secret-of-at-least-thirty-two-bytes!")

raw = JwtClaims({"sub": "42"}).token(signature).value()
print(raw)  # => "eyJhbGciOiJIUzI1NiIs..."

claims = JwtToken(raw, signature).claims()
print(claims.json())  # => {"sub": "42"}

The algorithm is an object, never a hardcoded string. Pick Hs256, Rs256, or Es256, or implement the Signature interface yourself.

Tokens That Expire

Wrap your claims in ExpiringClaims to add an exp claim. The lifetime is in seconds:

from elegant_jwt import ExpiringClaims, Hs256, JwtClaims

signature = Hs256("a-secret-of-at-least-thirty-two-bytes!")

token = ExpiringClaims(
    JwtClaims({"sub": "42"}),
    3600,
).token(signature)

print(token.expired())   # => False
print(token.validity())  # => 3600 (seconds left until expiration)

Refusing Expired Tokens

StrictToken is a decorator that refuses to give claims from an expired token. Use it wherever an expired token must be treated as an error:

from elegant_jwt import ExpiringClaims, Hs256, JwtClaims, JwtToken, StrictToken

signature = Hs256("a-secret-of-at-least-thirty-two-bytes!")
raw = ExpiringClaims(JwtClaims({"sub": "42"}), 3600).token(signature).value()

token = StrictToken(JwtToken(raw, signature))
token.claims()  # raises Exception once the token has expired

A strict token needs an exp claim to judge freshness, so create it with ExpiringClaims.

Stamping the Issuer

IssuedClaims adds iat (issued at) and iss (issuer) claims. Decorators stack, each one adding its own claims on top:

from elegant_jwt import ExpiringClaims, Hs256, IssuedClaims, JwtClaims

token = IssuedClaims(
    ExpiringClaims(
        JwtClaims({"sub": "42"}),
        3600,
    ),
    "my-service",
).token(Hs256("a-secret-of-at-least-thirty-two-bytes!"))

print(token.claims().json())
# => {"sub": "42", "exp": 1788094023, "iat": 1788090423, "iss": "my-service"}

Tokens Valid Only Later

NotBeforeClaims adds an nbf (not before) claim. The delay is in seconds from now; pyjwt refuses the token until that moment arrives:

from elegant_jwt import Hs256, JwtClaims, JwtToken, NotBeforeClaims

signature = Hs256("a-secret-of-at-least-thirty-two-bytes!")
raw = NotBeforeClaims(JwtClaims({"sub": "42"}), 300).token(signature).value()

JwtToken(raw, signature).claims()  # raises Exception for the next five minutes

Tokens for One Audience

A token that carries an aud (audience) claim is read back through AudienceSignature, a decorator that names the audience the reader expects. It refuses tokens addressed to anyone else, and tokens with no audience at all:

from elegant_jwt import AudienceSignature, Hs256, JwtClaims, JwtToken

signature = AudienceSignature(
    Hs256("a-secret-of-at-least-thirty-two-bytes!"),
    "ledger-service",
)
raw = JwtClaims({"sub": "42", "aud": "ledger-service"}).token(signature).value()

print(JwtToken(raw, signature).claims().json())
# => {"sub": "42", "aud": "ledger-service"}

The aud claim may also be a list; the token is accepted when the expected audience is one of its entries.

Asymmetric Algorithms

Rs256 and Es256 sign with a private key and verify with a public key, both in PEM format:

from elegant_jwt import JwtClaims, JwtToken, Rs256

signature = Rs256(private_pem, public_pem)

raw = JwtClaims({"sub": "42"}).token(signature).value()
claims = JwtToken(raw, signature).claims()

A service that only verifies tokens holds just the public key and never calls encoded.

Testing Without Waiting

Time is an input, not a hidden call. Every object that needs the current time accepts a Clock, so tests never sleep and never patch:

from elegant_jwt import Clock, ExpiringClaims, JwtClaims


class FrozenClock(Clock):
    def __init__(self, instant: int):
        self.instant = instant

    def moment(self) -> int:
        return self.instant


claims = ExpiringClaims(JwtClaims({"sub": "42"}), 60, FrozenClock(1000))
print(claims.json())  # => {"sub": "42", "exp": 1060}

Your Own Signature

Need a key from a JWKS endpoint, a vault, or a database? Implement the Signature interface and keep the policy (cache, retry, timeout) on your side; the library stays free of I/O:

from elegant_jwt import Signature


class VaultSignature(Signature):
    def __init__(self, vault: Vault):
        self.vault = vault

    def encoded(self, payload: dict) -> str:
        return Hs256(self.vault.secret()).encoded(payload)

    def decoded(self, raw: str, options: dict) -> dict:
        return Hs256(self.vault.secret()).decoded(raw, options)

Errors

Every failure raises a plain Exception with a human message, chained to the original cause:

try:
    JwtToken("not-a-token", signature).claims()
except Exception as trouble:
    print(trouble)  # => "The access token is not valid."

Design

• Every class is immutable; a change produces a new object.
• New behavior comes from decorators (StrictToken, ExpiringClaims, IssuedClaims, NotBeforeClaims, AudienceSignature), not from modification of existing classes.
• The library performs no network and no filesystem access.

Development

make unit     # tests with coverage
make black    # formatting
make flake8   # style
make ruff     # lint

How to Report Issues

Three rules, depending on what you found.

Enhancements. Open a GitHub issue and label it enhancement. Describe the desired behaviour and why it is useful; no code is required.

Bugs in code. Open a pull request, not an issue. The PR must contain a test that reproduces the bug and fails against the current code. Mark the test as disabled with pytest.mark.skip and a short reason, so CI stays green while the failing case is on record:

@pytest.mark.skip(reason="Reproduces #99, not fixed yet")
def test_keeps_claims_of_token_without_expiration():
    ...

The fix can arrive in the same PR or in a follow-up one, which removes the skip. Contributors without push rights fork the repository first.

Bugs outside code. If the bug cannot be reproduced with a test (documentation, packaging, CI configuration, and so on), open a GitHub issue and label it bug. Describe the expected and actual behaviour and how to observe it.