antlion-lacewing - v2.0.1
    Preparing search index...

    Verification

    Check Command
    Unit, compliance, conformance, properties, attacks, frameworks npm test
    Every requirement in tests/compliance/requirements.json has a passing test npm run compliance
    The built package through its exports map npm run test:dist
    Mutation testing npm run mutation
    Real Redis, Postgres, several processes, a Nimbus client npm run test:services

    test:services needs ANTLION_TEST_REDIS_URL, ANTLION_TEST_POSTGRES_URL, a JDK and the Nimbus jars. The CI services job sets them up.

    The compliance gate shows each requirement has a passing test, not that the test would fail if the check broke. Stryker mutates src/verify.ts, src/proof.ts and src/replay-store.ts one edit at a time and reruns the covering tests. On Node 26.11, Linux, Stryker 10.0.0, concurrency 4:

    File Mutants Killed Survived Score
    proof.ts 206 200 6 97.09%
    replay-store.ts 70 64 6 91.43%
    verify.ts 282 278 4 98.58%
    All 558 542 16 97.13%

    The first run scored 86.56% with 75 survivors. They were fixed by adding tests, not by excluding mutants. None of the 16 left changes anything a caller can observe:

    Where Mutation Why no test sees it
    proof.ts COMPACT_JWS drop the ^ or $ anchor Junk outside the segments is refused later as non-canonical base64url, same code
    proof.ts modulusBits drop bytes.length === 0 or bytes[0] === 0 An empty modulus never decodes; a leading zero gives minus infinity bits
    proof.ts checkKey check crv on RSA keys too No RSA algorithm names a curve
    proof.ts alg, iat drop the typeof guard includes() and Number.isFinite() refuse the same values
    proof.ts compactVerify drop algorithms: [alg] alg is already checked; jose uses the same value
    replay-store.ts (6) full sweep, or skip the reorder, on every call Only slower
    verify.ts refusalFor issue a nonce without checking nonces are on use_dpop_nonce only happens when they are
    verify.ts readRequest drop typeof request !== "object" A string has no method and is refused one line later
    verify.ts readHeaderValue label empty the label Only printed for a headers object readRequest already refused
    verify.ts Bearer check drop the ?. split(" ", 1) always has an element

    Stryker runs every test file except tests/types, which compile fixtures with tsc and can't type-check Stryker's instrumented source.

    tests/properties/ holds the fast-check suites. One changes a single character of a valid proof or token and requires a refusal, and checks the iat window against random clocks. The fuzz suite sends lone surrogates, NUL and CR/LF in both headers, correctly signed proofs with arbitrary header or payload bytes, megabyte values, and arbitrary methods and URLs. Every outcome is a typed AntlionError, and oversized values are refused before anything is decoded.

    tests/attacks/corpus.test.ts has one test per known attack: a stolen token with the attacker's key, replay, downgrade to Bearer, an unbound token, a self-issued token, htu prefix and host tricks, method swap, alg: none and HMAC with the public key, key swap, ath stripped or reused, pre-generated and stale proofs, duplicate headers, a JSON member named twice, a token presented as a proof, jku/x5u/x5c injection, and a private key in the header. Each refusal is also checked for leaks: no message contains part of the token or proof, and the wire refusal never names the internal code.

    • Golden vectors: RFC 9449's worked examples, and proofs made by python-cryptography and by Nimbus's DefaultDPoPProofFactory, regenerated by the scripts in tools/golden/.
    • oauth4webapi, live. A real HTTP server on 127.0.0.1 with a test token endpoint. oauth4webapi gets a bound token with ES256, PS256 and Ed25519 keys, is challenged for a nonce, retries, and reaches two routes. It is refused when its proof is replayed, when its clock is an hour slow, when another key presents its token, and when it sends the token as Bearer.
    • Nimbus, live. The same flow from Java with Nimbus OAuth 2.0 SDK 11.38.2, for all three key types.

    The OpenID Foundation's conformance suite has tested DPoP in FAPI 2.0 since 2023, but its plans certify authorization servers and clients. When it tests a client, it plays the resource server itself. As of October 2026 we found no plan that tests someone else's resource server.

    So Antlion has not been run through the FAPI conformance suite and claims no FAPI certification. It accepts only the FAPI 2.0 algorithm set, and the RFC 9449 resource-server requirements are covered section by section in tests/compliance/. If a resource-server plan appears, it will be run and its report kept here.

    • No external security review. Everything on this page is the maintainer checking their own work. The threat model is written for whoever does one.
    • Multi-process tests run on Linux only, because GitHub runs service containers only there. The stores' unit tests run on Linux, macOS and Windows.