Added #

• Project-level Universal Multi-Agent PQC Framework setup:
  • canonical framework guide in doc/UNIVERSAL_MULTI_AGENT_PQC_FRAMEWORK.md;
  • machine-readable manifest in tool/agent_framework/pqc_framework.yaml;
  • native thin wrappers for Codex, Claude Code, and Antigravity.
• example/main.dart now demonstrates ML-KEM shared-secret agreement, ML-DSA signing/verification, and an ML-DSA-signed ML-KEM-768 handshake transcript.
• doc/SERVERPOD_FLUTTER_GUIDE.md now covers ML-KEM + ML-DSA Serverpod/Flutter integration, strict byte contracts, generated model sketches, Flutter isolate guidance, and framework-driven implementation prompts.
• ML-DSA (FIPS 204) is now byte-exact against the official KAT corpus for ML-DSA-44, ML-DSA-65, and ML-DSA-87 across the full matrix of signing mode (deterministic, hedged) × implementation flavour (raw/internal, pure/external-with-context, hashed/HashML-DSA): 300/300 key generations and 1800/1800 signatures reproduced byte-for-byte, all verifying.
• External FIPS 204 API on MlDsa: generateKeyPair(params) (fresh randomness), sign/verify with a context string (≤ 255 bytes) and hedged-by-default signing, signDeterministic, and the internal/CAVP helpers generateKeyPairSeeded, signInternal, verifyInternal.
• HashML-DSA (hashSign/hashVerify) with the FIPS 204 §5.4 pre-hash: SHA-256 (ML-DSA-44), SHA-384 (ML-DSA-65), SHA-512 (ML-DSA-87), with DER OID domain separation.
• Vendored FIPS 180-4 SHA-2 (lib/src/common/sha2.dart): SHA-256/384/512, web-safe 64-bit (hi/lo pair) arithmetic, pinned by direct NIST vectors.
• Repo-local ML-DSA KAT corpus under test/data/MLDSA (18 .rsp files) with a discovered, deterministic runner test/mldsa_kat_test.dart; ML-KEM corpus moved to test/data/MLKEM.
• New focused tests: Appendix B zetas + negacyclic NTT property, rounding boundary cases, malformed-input/negative verification, external-API behavior, and direct SHA-2 vectors. All run on the VM, dart2js, and dart2wasm.
• SECURITY.md (vulnerability reporting / coordinated disclosure policy) and doc/FIPS_140_BOUNDARY.md (why algorithm conformance is not CMVP/FIPS 140 module validation), linked from every place the claim boundary is raised.

Changed #

• ML-KEM decapsulation hardening: the FIPS 203 output is now selected with a constant-time branchless mask — both K' and J(z || c) are always computed, so success vs. implicit-rejection no longer differs in control flow. Secret intermediates (m', K' || r', J(z || c), c', z) are zeroized in a finally block, and the KEM now reuses a cached Random.secure(). The 3000-vector ML-KEM KAT corpus (including invalid-ciphertext vectors) remains byte-exact.

• ML-DSA rejection samplers (RejNTTPoly, RejBoundedPoly, SampleInBall) now use an incremental SHAKE XOF (KeccakXof) instead of fixed buffers, so they cannot exhaust output during normal operation.

• ML-DSA packing preserves signed coefficient domains for s1/s2/t0/z instead of folding negatives into [0, q-1].

• _normExceeds (the ML-DSA norm gate) evaluates all 256 coefficients with no secret-dependent early exit, and uses only VM/web-portable arithmetic.

• Verification is total: MlDsa.verify/verifyInternal return false (never throw) for wrong pk/sig lengths, malformed hints, or over-long context.

• DilithiumParams exposes computed FIPS 204 Table 2 sizes (publicKeyBytes, secretKeyBytes, signatureBytes, plus per-poly sizes, lambda, securityCategory) as the single source of truth; the unused crhBytes constant was removed.

Fixed #

• RejBoundedPoly (ExpandS) for η=2: now accepts half-bytes < 15 mapping to 2 − (b mod 5) per FIPS 204, fixing the stream-consumption rate that made key generation diverge from the standard. This was the sole core defect.
• ML-DSA verification no longer used a 32-bit left shift (<< d) that overflowed on dart2js; it multiplies by 2^d so web results match the VM.

Security #

• Best-effort secret zeroization (lib/src/common/zeroize.dart) applied in finally blocks around key generation and signing intermediates. Dart cannot guarantee hard memory erasure; see doc/SECURITY_AUDIT.md for the boundary.

Notes #

• This repository continues to make no CMVP/FIPS 140 module validation claim. ML-DSA conformance evidence is the checked-in KAT corpus and regression suite.

Added #

• Expanded the OpenSSL interoperability harness from ML-KEM-768 only to ML-KEM-512, ML-KEM-768, and ML-KEM-1024.
• Added stronger OpenSSL interop checks:
  • Deterministic shared-seed key generation proves byte-identical public keys.
  • Public-key import/export round-trip proves raw wire encoding compatibility.
  • Invalid-ciphertext implicit rejection proves J(z || c) agreement.
  • Negative coverage confirms truncated OpenSSL public keys are rejected.
• Added VM + web round-trip tests so KEM keygen/encaps/decaps is exercised under the Dart VM, dart2js, and dart2wasm.
• Added FIPS 202 SHA-3/SHAKE known-answer coverage for the vendored Keccak implementation.

Changed #

• Removed the pointycastle runtime dependency by vendoring the FIPS 202 SHA3-256, SHA3-512, SHAKE128, and SHAKE256 implementation in-tree.
• Updated the README and OpenSSL interop documentation to describe the all-parameter-set A-G interop suite, OpenSSL >= 3.5 requirement, and zero-dependency pure-Dart package boundary.
• Updated CI to include web compiler testing and the expanded OpenSSL interoperability suite.

Added #

• Input validation for encapsulate() and decapsulate() per FIPS 203 §7.2/§7.3:
  • Public key length and modulus checks (non-canonical coefficient rejection via ByteEncode₁₂ ∘ ByteDecode₁₂ round-trip).
  • Secret key length and embedded H(pk) integrity check.
  • Ciphertext length check.
  • Pack.decodeSecretKey length guard.
• OpenSSL interoperability tool (tool/openssl_interop/): dart:ffi-based harness proving wire-level ML-KEM-768 compatibility with OpenSSL ≥ 3.5. Four-way test matrix (A/B/C/D) validates byte-identical shared secrets across implementations.
• CI workflows:
  • ci.yml: format check, static analysis, and full test suite (unit + 3000-vector KAT corpus) on every push/PR.
  • interop.yml: builds OpenSSL 4.0.0 from source (cached), runs the four interop tests on every push/PR.
• New tests:
  • kem_validation_test.dart: exercises all input validation paths across ML-KEM-512/768/1024.
  • keygen_derivation_test.dart: isolates FIPS 203 domain separation (G(d || k)) and matrix expansion ordering.
  • poly_test.dart: verifies barrettReduce returns canonical residues in [0, q).
• Documentation:
  • doc/MLKEM_TESTING.md: KAT file hashes, coverage boundaries, release-gate commands, and scoped claim boundary.
  • doc/OPENSSL_INTEROP.md: full interop guide with FFI bindings, versions, results, and use cases.
• Test hooks genMatrixEntryForTest / sampleNttForTest on Indcpa (internal, not exported).
• .pubignore to exclude dev-only files from the published package.

Changed #

• Naming conventions: renamed internal identifiers to idiomatic Dart lowerCamelCase (_H/_G/_J → _h/_g/_j; A_hat/t_hat/r_hat → aHat/tHat/rHat; etc.). No behavioral change.
• barrettReduce(): use const for compile-time constants and add a fallback res %= q guard for edge-case residues.
• Renamed test/kat_evaluator.dart → test/kat_evaluator_test.dart so dart test discovers it automatically.
• README rewritten with scoped validation claims, OpenSSL interop section, and corrected Markdown formatting.
• pubspec.yaml description: fixed typo ("Startss" → "Starts"), updated wording to "FIPS 203-aligned".

Removed #

• Unused Poly.montgomeryReduce() (the implementation uses Barrett reduction exclusively).

• Initial release of pqcrypto.
• Implements ML-KEM (Kyber) FIPS 203 standard.
• Supports ML-KEM-512, ML-KEM-768, and ML-KEM-1024.
• Pure Dart implementation with 3000/3000 NIST KAT vectors passing.
• Compatible with Flutter and Dart Web (Wasm/JS).