All notable changes to this project will be documented in this file.
The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.
Convention: Each release section uses emoji headers (## ✨ Added, ## 🔄 Changed, etc.)
so that .github/workflows/release.yml can extract them directly into GitHub Releases
without reformatting. The [Unreleased] section uses plain headers for drafting convenience;
rename the headers to emoji form when cutting a release.
- TBD
- TBD
- TBD
Release Date: June 14, 2026 Repository: bradh11/certmonitor
CertMonitor v0.4.0 brings post-quantum readiness to certificate and TLS monitoring. New opt-in validators report whether the TLS key exchange and a certificate's keys and signatures use post-quantum algorithms (hybrid or pure ML-KEM / ML-DSA / SLH-DSA), backed by an in-house Rust TLS 1.3 probe that reads the negotiated key-exchange group directly off the wire (something Python's ssl module does not expose). This release also formalizes the validator result envelope across every validator, upgrades pyo3 to resolve two security advisories, and ships a full documentation overhaul including a new Concepts section. Note the breaking change to cipher/TLS allow-list configuration described under Changed.
- Rust SPKI parser now recognizes post-quantum algorithm OIDs instead of collapsing them to
Unknown: ML-DSA-44/65/87 (FIPS 204, RFC 9881), all twelve SLH-DSA parameter sets (FIPS 205, RFC 9909), and the eighteen composite ML-DSA signature OIDs from draft-ietf-lamps-pq-composite-sigs-19.parse_public_key_infoandanalyze_chainreport e.g.algorithm: "ml-dsa-65"withsizeset to the subjectPublicKey bit length (PQ strength is judged by algorithm identity, not size). Dict shape is unchanged; existing RSA/EC behavior is regression-tested. Foundation for the upcoming PQ validators (#28, #29). certinfo.pq_algorithms(): exposes the Rust post-quantum algorithm registry (rust_certinfo/src/pq_algorithms.rs) to Python as[{"dotted", "name", "composite"}, ...], keeping the Rust table the single source of truth for PQ algorithm identity (#28, #30).- Rust
tls/module: TLS 1.3 handshake building blocks for the upcoming post-quantum key-exchange probe.key_exchange_groups.rs(IANA Supported Groups registry with PQ classification, contributor data file),records.rs(bounded record framing),handshake.rs(ClientHello builder with realistic extensions, ServerHello/HelloRetryRequest parser extracting the negotiated group). Pure parsing, no networking, no Python exposure yet; includes aparse_server_hellofuzz target (#28, #32). certinfo.probe_tls_handshake(host, port=443, timeout_ms=10000): opens a second TCP connection, sends a TLS 1.3 ClientHello offeringX25519MLKEM768with a real (NIST ACVP KAT) ML-KEM-768 key_share, reads the ServerHello, and reports the negotiated key-exchange group as{result, name, kind, is_pq, protocol, via_hello_retry_request}(or a structuredn/a/errordict: it never raises for network/protocol conditions). No crypto, no certificate validation; the socket work runs with the GIL released. Verified end-to-end against a live ML-KEM TLS 1.3 server. Foundation for thepq_key_exchangevalidator (#28, #33).pq_key_exchangevalidator (opt-in): reports the post-quantum posture of the TLS key exchange (the harvest-now-decrypt-later question). Returns{kem_id, kem_name, kem_kind, is_pq, is_valid}for a negotiated TLS 1.3 group (hybrid, e.g.X25519MLKEM768, and pure ML-KEM both count as PQ),is_valid: falsefor classical groups or TLS < 1.3, and a structured{error, message}on probe failure. Consumescertinfo.probe_tls_handshake; the probe's second connection is skipped entirely for TLS < 1.3. Not inDEFAULT_VALIDATORS(#28, #34).pq_chainvalidator (opt-in): walks the presented certificate chain and reports per-certificate post-quantum posture (certslist, matching thechainvalidator's shape):{position, role, key_algorithm, key_is_pq, signature_algorithm_oid, signature_is_pq, is_pq}plus a{leaf_pq, intermediate_pq, root_pq}summary. A cert counts as PQ when its key or signature algorithm is post-quantum (standalone or composite).is_validdefaults to "leaf key is PQ" (the part the operator controls);require_full_chain: truedemands the whole chain. Classical public roots are expected during migration and documented as such. Not inDEFAULT_VALIDATORS.pq_signaturevalidator (opt-in): judges the leaf certificate's post-quantum posture:{key_algorithm, key_is_pq, signature_algorithm_oid, signature_is_pq, is_hybrid_composite, is_pq, is_valid}(per-cert field names matchpq_chain).is_validdefaults to "leaf key is PQ";require_pq_signature: trueadditionally demands a PQ signature from the CA. Works on every supported interpreter via a new leaf-only analysis fallback: when the chain cannot be retrieved (Python 3.8/3.9 or a chain failure),cert_data["leaf_analysis"]is populated from the leaf DER. Not inDEFAULT_VALIDATORS.
- Added
make version(reports the current version frompyproject.tomlandCargo.toml) andmake version.patch/version.minor/version.majorto bump both in lockstep, modeled onuv versionbut spanning the Python and Rust artifacts. make fuzzandmake fuzz-longnow offer to install missing prerequisites (cargo-fuzz, nightly Rust toolchain) interactively when run from a terminal, instead of just printing the install command and exiting. CI and other non-interactive contexts retain the existing print-and-exit behavior via a[ -t 0 ]stdin check.key_infovalidator now recognizes post-quantum keys: certificates using ML-DSA, SLH-DSA, or composite ML-DSA algorithms returnis_valid: trueinstead of the misleadingis_valid: null. PQ strength is judged by algorithm identity (the registry above); RSA/EC behavior is unchanged. (Unknown/undeterminable keys now fail closed withis_valid: falseper the result envelope, see Fixed.) (#30)- Completed the result-envelope modernization (#46): every validator now declares a per-validator
TypedDictextendingValidationResultand annotates itsvalidate()return type, so mypy enforces the envelope contract across the whole validator suite. No runtime behavior change: results remain plain dicts. - Improved discoverability: sharpened the package
descriptionand added PyPIkeywordsand troveclassifiers(including post-quantum / ML-KEM / ML-DSA and Topic :: Security :: Cryptography), and added a "Post-Quantum Readiness" section plus the threepq_*validators to the README. - Validator dispatcher generalized to a declared data-source model: validators name the data they need via
requires(e.g.("cert_data",),("cipher_info", "tls_probe")); the dispatcher fetches and memoizes each source once per scan and injects them in declaration order. Existing validators are unchanged. A source-fetch failure now produces a uniform structured error result for every dependent validator, fixing a latent bug where a cipher-info fetch error silently dropped cipher validators from the results dict. - The cipher-suite and TLS-version allow-lists now live in their validators as per-call user arguments instead of mutable module globals.
weak_cipheracceptsallowed_cipher_suitesandtls_versionacceptsallowed_tls_versions(each defaulting to the built-in set), matching the configurable pattern already used bychain. Breaking:certmonitor.cipher_algorithms.ALLOWED_CIPHER_SUITES,ALLOWED_TLS_VERSIONS, andupdate_allowed_lists()have been removed; configure viavalidator_argsinstead.parse_cipher_suite/update_algorithmsare unchanged. This also removes the global mutable state that previously allowed configuration to leak between runs. Theweak_cipherdefault now includes the three DHE-RSA AEAD suites it was missing from Mozilla "Intermediate". - Upgraded
pyo3from 0.24 to 0.29, resolving two advisories affecting the shipped extension: RUSTSEC-2026-0176 (out-of-bounds read innth/nth_backforPyList/PyTupleiterators) and RUSTSEC-2026-0177 (missingSyncbound onPyCFunction::new_closureclosures). The only source change required wasPython::allow_threadstoPython::detach(a pyo3 rename) in the TLS probe's GIL-release path.
key_infono longer reports every EC certificate asis_valid: false. The Rust parser emitted the EC curve as an OID (e.g.1.2.840.10045.3.1.7) while the validator compared against curve short names, so the strong-curve check never matched.parse_public_key_infonow reports the curve by its documented short name (e.g.secp256r1), falling back to the OID string for curves outside the known table. Latent since the in-tree parser rewrite (v0.3.0); curve identity stays in the Rust registry (#48).weak_cipherno longer reports every TLS 1.3 connection as using a disallowed cipher. The allowed-cipher list held only TLS 1.2 (OpenSSL-style) names whileALLOWED_TLS_VERSIONSpermits TLS 1.3, so a TLS 1.3 handshake (the modern default) failed on its strongest cipher. The three standard TLS 1.3 suites (TLS_AES_128_GCM_SHA256,TLS_AES_256_GCM_SHA384,TLS_CHACHA20_POLY1305_SHA256) are now allowed (#50).- Fixed test state leakage where
update_allowed_listsmutated the module-global allow-lists without restoration, letting one test's custom lists bleed into later tests in the same run. - Brought four default validators into conformance with the standard result envelope (
is_validstrictbool;reasonpresent wheneveris_validisFalse):expiration,root_certificate, andsensitive_datesetis_valid: falseon failure but carried the explanation only inwarnings, so readingresult["reason"]on a failure raisedKeyError; each now adds a concisereason.key_inforeturnedis_valid: nullfor unrecognized algorithms or missing size/curve. It now fails closed (is_valid: falsewith areasondistinguishing "cannot determine" from "recognized but weak"). Verified against live hosts with zero envelope violations. - TLS probe ClientHello random is now guaranteed to vary between calls. The previous time-plus-stack-address seed could collide when the clock did not advance between two calls (observed on macOS's coarse
SystemTimeresolution), producing identical "random" bytes and a flaky test; a monotonic per-call counter now guarantees distinct values (#33).
Comprehensive documentation is available at certmonitor.readthedocs.io.
Tested with Python 3.8 through 3.13 with 99% code coverage across all supported versions.
This project is licensed under the MIT License. See the LICENSE file for details.
Full Changelog: https://github.com/bradh11/certmonitor/compare/v0.3.0...v0.4.0
Release Date: April 15, 2026 Repository: bradh11/certmonitor
CertMonitor v0.3.0 is a zero-dependency milestone. The Rust extension's entire X.509 / DER parser is now written in-house against the Rust standard library, with no third-party parsing crates in the runtime dependency tree. The Rust dependency count drops from 48 crates to 20, with every remaining crate being pyo3 or a build-time helper.
The parser is annotated #![forbid(unsafe_code)] at the crate root, returns Result on every code path (no panics on malformed input), and has been fuzz-tested against 1.7 billion adversarial byte sequences with zero crashes.
This release also ships the chain validator for structural inspection of TLS certificate chains, and fixes two latent bugs in the public key info output.
chainvalidator (#14): structural validation of the full TLS certificate chain. Flags missing intermediates, out-of-order chains, expired members, weak signature algorithms (SHA-1, MD5), non-CA intermediates, and unexpected self-signed leaves. Registered but disabled by default: opt in viaenabled_validators=["chain"]orENABLED_VALIDATORS=...,chain. Chain retrieval requires Python 3.10+; returns a clear error on older interpreters.certinfo.analyze_chain(Rust): parses a full DER chain in a single PyO3 call and returns per-cert details plus subject/issuer and SKI/AKI linkage.SSLHandler.fetch_raw_certnow additionally returnschain_derandchain_error, populated viaSSLSocket.get_verified_chain()on Python 3.13+ and the stable_sslobj.get_unverified_chain()fallback on 3.10-3.12.- In-tree DER / X.509 parser (#22): a strict-DER, no-
unsafe, panic-free parser underrust_certinfo/src/{der,x509}/. Theder/layer (TLV reader, OID decoder, time parser, string decoders) is reusable for future ASN.1-based capabilities. Thex509/layer (Certificate, Name, SPKI, AlgorithmIdentifier, Extensions) composes those primitives into RFC 5280 structures. - Fuzz harness (#25):
make fuzz(60-second smoke run) andmake fuzz-long(1-hour soak) Makefile targets. Pre-release soak run results: 1.7 billion iterations, 310 code-coverage points, 503 libfuzzer features explored, zero crashes. Requires nightly Rust +cargo-fuzz. Manual pre-release gate, not CI. - 130-cert real-world corpus (
tests/test_certinfo_corpus.py): snapshot tests run every publiccertinfoentry point against captured certs from 101 production hosts on every CI run. - 56 in-module Rust unit tests covering DER primitives, OID round-trips, time parsing, Name/RDN walking, SPKI dispatch, and extension parsing.
pythonCargo feature on thecertinfocrate (default on). Disabling it drops the PyO3 layer entirely and builds only the pure-Rust parser core, used by the fuzz crate.scripts/bench_chain.py: opt-in benchmark with a microbench ofanalyze_chain(~400 µs/call) and a 101-host concurrent pipeline test.
- Zero non-pyo3 Rust dependencies.
x509-parserandbase64are gone. The Rust dependency tree shrinks from 48 crates to 20: every remaining crate ispyo3or a build-time helper.cargo auditsurface drops accordingly. Cargo.tomlcrate-type is now["cdylib", "rlib"]. Thecdylibis the same Python wheel target;rliblets the fuzz crate link the parser as a normal Rust library. No published-wheel surface change.certinfo::Certificate::from_derandcertinfo::ParseErrorare nowpubat the crate root for use by the fuzz crate and future in-tree Rust consumers. The PyO3 boundary and Python-facing API are unchanged.
- EC
curvefield now correctly contains the curve OID. Previous builds emitted the algorithm OID1.2.840.10045.2.1(id-ecPublicKey) in the field literally namedcurve. The new parser extracts the curve OID fromalgorithm.parameters:1.2.840.10045.3.1.7for P-256,1.3.132.0.34for P-384,1.3.132.0.35for P-521. Visible behavior change for any caller readingpublic_key_info["curve"]. - RSA modulus bit length is no longer over-counted by 8 bits. Previous builds reported RSA-2048 / 3072 / 4096 keys as 2056 / 3080 / 4104 due to including the DER-mandated leading-zero sign byte. The new parser reports the canonical 2048 / 3072 / 4096. Visible in
public_key_info["size"].
- New per-validator docs page:
docs/validators/chain.md. - README: new "Why Trust CertMonitor" section with fuzz results, zero-dep guarantee,
forbid(unsafe_code), and coverage numbers. - Comprehensive documentation at certmonitor.readthedocs.io.
Tested with Python 3.8 through 3.13 with 99% code coverage across all supported versions. The chain validator requires Python 3.10+ for chain retrieval; all other features work on 3.8+.
This project is licensed under the MIT License. See the LICENSE file for details.
Full Changelog: https://github.com/bradh11/certmonitor/compare/v0.2.0...v0.3.0
Release Date: April 13, 2026 Repository: bradh11/certmonitor
CertMonitor v0.2.0 overhauls how validators receive arguments. New validators can now declare their user arguments directly on the validate() method signature: the dispatcher discovers them automatically and no core changes are needed. As part of the same effort, the sensitive_date validator (which has been sitting on develop since #15 back in June 2025) finally makes it into a release, and gets ergonomic input forms, a structured match field, and structured error handling along the way.
This is a minor version bump to reflect the scale of the changes, not because of any hard break in the public API. Existing callers using validator_args={"subject_alt_names": [...]} still work with a DeprecationWarning, and no validator output shape has changed for users.
- Dynamic validator argument dispatch (#18): validators declare their user-configurable arguments directly on the
validate()method signature, andCertMonitor.validate(validator_args=...)discovers them automatically. New validators get argument passing for free: zero core changes needed. CertMonitor.describe_validators(): new introspection helper that returns every registered validator's name, docstring, and argument schema (name, annotation, default). Useful for building CLI--helppages, config validators, or dashboards.sensitive_datevalidator finally ships: flags certificates that expire on weekends, leap days, or user-specified dates (e.g. Black Friday, Cyber Monday, go-live dates).sensitive_dateinput ergonomics: thedatesargument acceptsSensitiveDatenamed tuples, plaindate/datetimevalues, ISO 8601 strings ("2025-12-25"), or(name, date)tuples, all mixable in a single call. No need to importSensitiveDatefrom a deeply nested module path just to pass a list of blackout dates.sensitive_date_matchesstructured field: matching sensitive dates are surfaced as a machine-readable list of{"name", "date"}entries in addition to the existing human-readablewarningsstrings.- Weekend / leap-day warning strings: when the
sensitive_datevalidator flags a weekend or leap-day expiry, a human-readable warning line is now emitted alongside the existing boolean fields, so log output is self-explanatory whenis_validis false. - Shared
parse_not_afterhelper (certmonitor/validators/_utils.py): centralizes thenotAfterformat string shared byexpirationandsensitive_date.
- Validator author contract: user arguments on
validate()must be keyword-only, type-annotated, and have a default value. Enforcement runs inBaseCertValidator/BaseCipherValidator__init_subclass__at import time, so a malformed validator raisesTypeErrorthe moment its module is imported. No user-facing impact: every built-in validator conforms, and the dispatcher continues to accept the pre-0.2.0validator_argscall style via a deprecation shim. subject_alt_namesandsensitive_datesignatures migrated to keyword-only user arguments (alternate_names=...,dates=...). Existing users ofmonitor.validate(validator_args={...})are unaffected; callers invoking the validator classes directly with positional arguments need the keyword form.validator_argscanonical form is now a nested dict:validator_args={"subject_alt_names": {"alternate_names": [...]}}. The pre-0.2.0 bare-list form still works and is transparently rewritten by the dispatcher (with aDeprecationWarning), so no user code needs to change immediately.sensitive_dateerror handling: malformeddatesinput (wrong type, invalid ISO string, bad tuple shape) now returns a structured error dict instead of raisingTypeError, matching the rest of the validator suite.expirationvalidator: now uses the sharedparse_not_afterhelper; behavior unchanged.mkdocs.yml: added the previously-missingSensitiveDatenav entry so the validator's auto-generated reference page is reachable.docs/usage/validator_args.md: rewritten to document the canonical nested-dict form,describe_validators(), the bare-list deprecation, and a workedsensitive_dateexample showing all four input forms.- Rust toolchain floor moved to
rustc >= 1.88.0(transitively via thetime 0.3.47security bump, see below). Affects contributors and source builds only: published wheels are unaffected.
- Bare-list shorthand for single-argument validators (
validator_args={"subject_alt_names": [...]}) still works but now emits aDeprecationWarning. Migrate to the canonical nested-dict form. Scheduled for removal in a future release.
- RUSTSEC-2026-0009: bumped the
timecrate from0.3.41to0.3.47(transitively viax509-parser) to address the denial-of-service-via-stack-exhaustion advisory.
subject_alt_namescore dispatch: the hardcodedif validator.name == "subject_alt_names"special case incore.validate()is gone, replaced with the generic argument-resolution helper used by every validator.CHANGELOG.md: backfilled the missing[0.1.4]section from the published release notes so the historical record is complete.
Comprehensive documentation is available at certmonitor.readthedocs.io.
Tested with Python 3.8 and above with 98%+ code coverage across all supported versions.
This project is licensed under the MIT License. See the LICENSE file for details.
Full Changelog: https://github.com/bradh11/certmonitor/compare/v0.1.4...v0.2.0
Focus: test coverage and CI optimization (#16).
- Achieved 99% test coverage (up from 95%) with comprehensive edge case testing.
- Instance convenience methods for improved developer experience:
monitor.get_enabled_validators(): get validators enabled for this specific instance.monitor.list_validators(): get all available validators.
- Enhanced test suite with 323 tests covering all edge cases.
- Streamlined security scanning: removed heavy semgrep dependency, kept focused bandit scanning.
- Improved test descriptions: removed line-number references for maintainable, functionality-focused tests.
- Enhanced validator configuration: proper distinction between empty lists vs config defaults.
- Default validator behavior:
enabled_validators=[]now properly means "no validators", vsNonemeaning "use defaults" (see #16). - Configuration environment handling: proper string parsing for the
ENABLED_VALIDATORSenvironment variable. - Test coverage gaps: targeted testing for previously uncovered edge cases: SSL handler retry exception scenarios, certificate parsing fallback mechanisms, handler-None conditions in raw data operations, and public key parsing error paths.
- Comprehensive GitHub workflows and templates with develop/main branch strategy
- Enhanced type hints throughout entire codebase (zero mypy errors)
- Modularized test suite for better maintainability
- ReadTheDocs integration with proper configuration
- Consolidated CI/CD pipeline with conditional job execution
- Unified Makefile commands for Python + Rust development workflow
make format- Format both Python and Rust codemake format-check- Check formatting for both languagesmake lint- Lint both Python and Rust codemake security- Run security vulnerability check (cargo audit)- Individual language commands:
python-format,python-lint,rust-format,rust-lint - Enhanced
make testwith 9-step CI-equivalent comprehensive testing including security checks - Improved
make helpwith clear categorized command documentation
- Improved code organization and structure
- Enhanced documentation and contributing guidelines
- Updated all workflows to use develop/main branch strategy
- Removed redundant CI configurations (quality.yml, security.yml, rust.yml, docs.yml)
- Streamlined dependency management (removed Dependabot for stdlib-only project)
- Enhanced local development experience with unified format/lint commands
- Makefile now provides comprehensive Python + Rust development workflow
- All mypy type errors across 20 source files using proper type annotations and runtime checking
- Import sorting and code quality issues
- SSL handler connection logic and check_connection() functionality
- Python 3.8 compatibility issues (datetime.UTC → datetime.timezone.utc)
- CI workflow syntax errors and redundant documentation building
- Fixed incorrect reporting on root certificate validation
- Security vulnerability RUSTSEC-2025-0020 by upgrading PyO3 from 0.20.0 to 0.24.1
- Rust code compatibility with PyO3 0.24.x API changes (updated module binding syntax)
- ABI3 support for Python wheels for better compatibility
- Improved wheel building process for cross-version compatibility
Note: This release has been deprecated. Please use v0.1.2 or later.
- Python 3.13 support in testing matrix
- Trusted publisher configuration for PyPI releases
- Enhanced CI/CD pipeline with proper release permissions
- Updated package structure (moved Rust library outside of Python package)
- CI workflow issues and version bumping process
Note: This release has been deprecated due to Rust build failures. Please use v0.1.2 or later.
- Initial release of certmonitor
- Certificate validation and monitoring capabilities
- SSL/TLS certificate analysis
- SSH certificate support
- Hybrid Python-Rust implementation for performance
- Virtual environment setup in CI publish job
- Rust build failures preventing proper installation