github.com/ludo-technologies/pyscn

What's Changed

• feat(complexity): add per-function SLOC metric and long-function warnings (#551) by @DaisukeYoda in #660
• fix(analyzer): stop cognitive complexity from crossing scope boundaries (#658) by @DaisukeYoda in #661
• fix: make dependency analysis scale by @M-Hassan-Raza in #695
• fix: derive single-responsibility concerns from fan-out only by @DaisukeYoda in #700
• fix(config): stop config discovery at the target repository root by @DaisukeYoda in #701
• docs(ci): record the tag ruleset as what gates registry publishing by @DaisukeYoda in #684
• fix(parser): reject legacy syntax no Python 3 release accepts by @DaisukeYoda in #702
• build(deps): bump the go-dependencies group with 3 updates by @dependabot[bot] in #699
• fix(check): fail the gate on files that cannot be parsed by @DaisukeYoda in #703
• fix: --min-complexity filters the analysed function set, not just the report:... by @yhay81 in #697
• chore: bump polyscan core to v0.2.2 (APTED/LSH optimizations) by @DaisukeYoda in #707
• chore: translate Japanese code comments to English by @DaisukeYoda in #713
• fix: gate all analyzed functions by @sanmaxdev in #712
• fix: make class ranking ties deterministic by @sanmaxdev in #710
• migrate: adopt core parse-error penalty by @DaisukeYoda in #714
• docs: update CHANGELOG for v1.29.1 by @DaisukeYoda in #715

Full Changelog: v1.29.0...v1.29.1

Added

• Per-function SLOC metric and long-function warnings in check and the HTML report (#660)

Changed

• Adopt polyscan core parse-error penalty (#714)
• Bump polyscan core to v0.2.2 with APTED/LSH optimizations (#707)
• Make dependency analysis scale on large graphs (#695)
• Translate remaining Japanese code comments to English (#713)

Fixed

• Stop cognitive complexity from crossing scope boundaries (#661)
• Gate all analyzed functions in check, not just the reported ones (#712)
• Make class ranking ties deterministic (#710)
• Fail the check gate on files that cannot be parsed and gate clone detection (#703)
• Reject legacy Python 2 syntax that no Python 3 release accepts (#702)
• Discover config from the analyzed target and stop at its repository root (#701)
• Derive single-responsibility concerns from fan-out only (#700)
• Keep --min-complexity from shrinking the analyzed function set and health score (#697)

What's Changed

• fix(domain): preserve architecture suggestion identity by @yhay81 in #652
• docs: serve documentation at docs.codescan.dev by @DaisukeYoda in #685
• Add per-module quality hotspot metrics by @M-Hassan-Raza in #686
• Add per-directory complexity rollups by @M-Hassan-Raza in #687
• fix(domain): add dedup and file_path/start_line for architecture suggestions (#647) by @DaisukeYoda in #648
• docs: update CHANGELOG for v1.29.0 by @DaisukeYoda in #688

Full Changelog: v1.28.0...v1.29.0

Added

• Per-module quality hotspot metrics in analyze reports (#686)
• Per-directory complexity rollups in analyze reports (#687)

Changed

• Serve documentation at docs.codescan.dev (#685)

Fixed

• Dedupe architecture suggestions and add file_path/start_line (#648)

What's Changed

• docs: point cross-language link to polyscan instead of archived jscan by @DaisukeYoda in #681
• feat(mcp): list pyscn in the MCP Registry and Glama by @DaisukeYoda in #682
• docs: update CHANGELOG for v1.28.0 by @DaisukeYoda in #683

Full Changelog: v1.27.0...v1.28.0

Added

• List the pyscn MCP server in the official MCP Registry and Glama (#682)

Changed

• Point the cross-language link to polyscan instead of the archived jscan (#681)

What's Changed

• migrate: adopt core/clone from polyscan (grouping, dedup, similarity gates, classifier) by @DaisukeYoda in #664
• test: cover mock data use case by @sanmaxdev in #665
• fix(analyzer): exclude Cython primitive types from CBO dependent classes (#644) by @DaisukeYoda in #646
• docs(website): document Agent Skills as a recommended AI integration by @DaisukeYoda in #668
• migrate: adopt core APTED from polyscan by @DaisukeYoda in #669
• migrate: adopt core CFG analyses by @DaisukeYoda in #670
• docs(website): redesign homepage with CFG diagram hero by @DaisukeYoda in #671
• migrate: adopt core graph analyses by @DaisukeYoda in #672
• migrate: adopt core LSH, LCOM, DFA, and semantic evidence from polyscan by @DaisukeYoda in #673
• migrate: adopt core scoring and grade mapping from polyscan by @DaisukeYoda in #674
• Fix module-qualified enum coupling by @ColumbusLabs in #675
• feat(report): link reports and CLI output back to the GitHub repo by @DaisukeYoda in #679
• docs: update CHANGELOG for v1.27.0 by @DaisukeYoda in #680

New Contributors

• @ColumbusLabs made their first contribution in #675

Full Changelog: v1.26.4...v1.27.0

Added

• Link HTML reports and CLI output back to the pyscn GitHub repository (#679)

Changed

• Adopt polyscan core clone grouping, dedup, similarity gates, and classification (#664)
• Adopt polyscan core APTED tree edit distance (#669)
• Adopt polyscan core CFG analyses (#670)
• Adopt polyscan core graph analyses (#672)
• Adopt polyscan core LSH, LCOM, DFA, and semantic evidence (#673)
• Adopt polyscan core scoring and grade mapping (#674)
• Redesign the website homepage with a CFG diagram hero (#671)
• Document Agent Skills as a recommended AI integration on the website (#668)

Fixed

• Exclude Cython primitive types from CBO dependent classes (#646)
• Fix module-qualified enum coupling in CBO (#675)

What's Changed

• fix(analyzer): exclude same-file top-level peer classes from CBO (#637) by @DaisukeYoda in #639
• fix(mcp): propagate analyze recursive override by @yhay81 in #656
• docs(mcp): correct configuration discovery by @yhay81 in #650
• fix(mcp): publish implemented output controls by @yhay81 in #651
• fix(check): fail on malformed discovered .pyscn.toml by @yhay81 in #654
• refactor!: unify request bool override semantics by @yhay81 in #649
• fix(architecture): default StrictMode to false when auto-detecting layers without user config (#659) by @DaisukeYoda in #662
• docs: update CHANGELOG for v1.26.4 by @DaisukeYoda in #663

Full Changelog: v1.26.3...v1.26.4

Changed

• Unify request bool override semantics with sparse optional fields (#649)
• Correct MCP configuration discovery documentation (#650)

Fixed

• Default architecture strict mode to false when auto-detecting layers without user config (#662)
• Propagate recursive override in MCP analyze tool (#656)
• Fail check command on malformed discovered .pyscn.toml (#654)
• Publish only implemented output controls in MCP tool schemas (#651)
• Exclude same-file top-level peer classes from CBO coupling (#639)

What's Changed

• fix: ignore module constants in CBO dependencies by @sanmaxdev in #638
• feat(complexity): report pre-filter function count in summary (#597) by @DaisukeYoda in #635
• docs: update CHANGELOG for v1.26.3 by @DaisukeYoda in #645

Full Changelog: v1.26.2...v1.26.3

Added

• Report pre-filter function count (functions_total) in complexity summary (#635)

Fixed

• Ignore imported module constants (e.g. re.DOTALL) in CBO dependencies (#638)

What's Changed

• feat(analyzer): exclude abstract methods from LCOM4 grouping (#559) by @DaisukeYoda in #625
• fix(app): apply exclude patterns when all paths are pre-resolved files (#590) by @DaisukeYoda in #629
• test(analyzer): pin LCOM4 InstanceVariables regression for issue #627 by @DaisukeYoda in #630
• fix(cbo): exclude annotation-only deps under PEP 563 future annotations by @vaishnavidesai09 in #634
• fix: report only detected clone fragments by @sanmaxdev in #633
• fix(config): honor explicit threshold overrides matching defaults by @DaisukeYoda in #560
• refactor(config): unify MergeConfig on zero-value sentinels, remove FlagTracker by @DaisukeYoda in #640
• docs: update CHANGELOG for v1.26.2 by @DaisukeYoda in #642

New Contributors

• @vaishnavidesai09 made their first contribution in #634

Full Changelog: v1.26.1...v1.26.2

Changed

• Unify config merge logic on zero-value sentinels and remove the legacy FlagTracker mechanism (#640)

Fixed

• Exclude abstract methods from LCOM4 grouping so abstract base classes aren't penalized (#559)
• Apply exclude patterns when all input paths are pre-resolved files (#590)
• Exclude annotation-only dependencies from CBO under PEP 563 future annotations (#628)
• Report only clone fragments referenced by returned clone pairs/groups (#631)
• Honor explicit complexity threshold overrides that equal the default (#553)

What's Changed

• docs(readme): refresh Features section with five analysis angles by @DaisukeYoda in #621
• docs(readme): clarify DI anti-pattern detection is opt-in by @DaisukeYoda in #622
• feat(service): fix test-file exclude patterns not matching nested paths (#532) by @DaisukeYoda in #623
• docs: update CHANGELOG for v1.26.1 by @DaisukeYoda in #624

Full Changelog: v1.26.0...v1.26.1

Fixed

• Test-file exclude patterns (test_*.py, *_test.py) not matching files in nested directories (#532)

Changed

• Refresh README Features section with five analysis angles (#621)
• Clarify DI anti-pattern detection is opt-in in README (#622)

What's Changed

• fix(cbo): don't count a class defined within a class as coupling by @StressTestor in #562
• feat(claude-plugin): split single skill into focused agent skills by @DaisukeYoda in #618
• docs: recommend Agent Skills install via add-skills, make skills CLI-first by @DaisukeYoda in #619
• docs: update CHANGELOG for v1.26.0 by @DaisukeYoda in #620

Full Changelog: v1.25.0...v1.26.0

Added

• Add cli-analysis Agent Skill for CI setup and report generation without MCP

Changed

• Split the plugin's single skill into health-check, refactoring, and architecture-review skills for better trigger accuracy

Fixed

• Move the plugin skill into skills/<name>/SKILL.md so Claude Code discovers it; the previous flat skills/SKILL.md layout was never loaded
• Document the plugin install path on the website MCP page and correct the Claude Code setup instructions (claude mcp add, not claude_desktop_config.json)
• Don't count a class defined within a class as CBO coupling (#562)

What's Changed

• feat(analyzer): build CommunityGraph from DependencyGraph (issue #579) by @DaisukeYoda in #587
• feat(analyzer): implement deterministic Leiden community detection (issue #580) by @DaisukeYoda in #589
• [communities] Domain types, service & use-case wiring for community analysis by @DaisukeYoda in #591
• fix: collapse enum member CBO dependencies by @sanmaxdev in #594
• [communities] CLI --select communities + [communities] config (opt-in) by @DaisukeYoda in #592
• feat(service): community analysis JSON formatter (#584) by @DaisukeYoda in #596
• [communities] Text/HTML/YAML/CSV/DOT summary output by @DaisukeYoda in #598
• [communities] Test suite & documentation for community detection by @DaisukeYoda in #599
• feat(communities): add package mismatch scoring by @DaisukeYoda in #607
• feat(communities): layer mismatch scoring by @DaisukeYoda in #608
• docs: clarify architecture defaults by @sanmaxdev in #606
• feat(communities): community risk score + Health Score integration by @DaisukeYoda in #609
• feat(communities): interactive macro-architecture graph in HTML report by @DaisukeYoda in #610
• feat(communities): AI-agent context map + MCP docs by @DaisukeYoda in #611
• Graduate community detection default analyze by @LunaHeys in #612
• [codex] fix src-layout community graph edges by @LunaHeys in #613
• fix: drop clone groups without backing pairs by @sanmaxdev in #615
• chore(deps): bump the go-dependencies group with 4 updates by @dependabot[bot] in #614
• docs: add multilingual README translations by @DaisukeYoda in #616
• docs: update CHANGELOG for v1.25.0 by @DaisukeYoda in #617

New Contributors

• @sanmaxdev made their first contribution in #594

Full Changelog: v1.24.3...v1.25.0

Added

• Build a community graph from the dependency graph (#579)
• Deterministic Leiden community detection (#580)
• Domain types, service, and use-case wiring for community analysis (#582)
• CLI --select communities and [communities] config, opt-in (#583)
• Community analysis JSON formatter (#584)
• Text/HTML/YAML/CSV/DOT community summary output (#585)
• Package mismatch scoring (#600)
• Layer mismatch scoring (#601)
• Community risk score with Health Score integration (#602)
• Interactive macro-architecture graph in the HTML report (#603)
• AI-agent context map and MCP documentation (#604)

Changed

• Run module community detection in default pyscn analyze; opt out with --skip-communities or [communities] enabled = false (#605)
• Bump Go dependencies (#614)

Fixed

• Drop clone groups without backing pairs (#615)
• Resolve src-layout community graph imports (#613)
• Collapse enum member CBO dependencies (#594)

Changed

• Integrate clone statistics into detector result types (#561)

Fixed

• Count self.<property> reads as method-call edges in LCOM (#558)

Changed

• Use cognitive complexity metrics for complexity risk assessment (#552)
• Change duplication metric from group density to fragment ratio (#555)

Fixed

• Prefer highest-similarity clone type for group classification (#556)
• Group namespace import members so CBO isn't inflated by member explosion (#554)

Fixed

• Populate fragment hash in clone output (#549)
• Suppress clone groups covered by another group's larger windows (#548)
• Calibrate progress bar with actual task timings (#541)

Changed

• Speed up clone detection by reusing APTED DP matrices and parallelizing pair comparison (#538)

Fixed

• Preserve every elif in a chain so complexity isn't undercounted (#537)

Fixed

• Stop counting imported function calls as class dependencies in CBO analysis (#535)
• Penalize Type-4 similarity for disjoint string literal vocabularies (#534)

Changed

• Soften CBO coupling scoring calibration (#528)
• Soften LCOM cohesion scoring calibration (#530)

Added

• Architecture style presets (layered, hexagonal, clean, mvc) for dependency rule configuration (#520, #521, #523, #526)

Fixed

• Merge overlapping unreachable findings in dead code detection (#519)

Fixed

• Clear dependency CVEs and suppress math/rand false positive in security report (#513)
• Exclude lazy function-body imports from circular dependency detection while retaining them as edges (#514)

Changed

• Consolidate cognitive complexity into the analyzer layer (#515)

Fixed

• Register plugin MCP server by moving the manifest into a nested .claude-plugin/ directory (#509)

Changed

• Bump Go dependencies (#504)
• Upgrade golangci-lint-action to v7 (#507)

Fixed

• Collect module imports from else/except/finally branches with branch-aware TYPE_CHECKING filtering (#502)
• Populate per-fragment id and type in clone group output (#503)
• Reuse detector-owned clone feature cache when building LSH MinHash signatures (#505)

Fixed

• Propagate nested function CFGs and reset builder state between builds (#497)
• Fix DFA compound-statement use contracts: match pattern bindings, definition scopes, and parameter/function-header uses (#498)
• Skip trailing-semicolon empty statements in dead code detection (#500)

Fixed

• Fix analyzer import and report contracts (dependency traversal, module imports, comprehension fields) (#495)
• Escape HTML report output to prevent markup injection (#495)

Fixed

• Exclude Django migration files from analysis defaults (#489)
• Compute TotalClones from clone group membership, not all fragments (#487)

Fixed

• Preserve clone group type classification (#484)
• Preserve skip_docstrings config in clone detection (#483)
• Detect external ctypes fields and broaden ctypes base detection in LCOM (#480)

Fixed

• Resolve CBO dependency identity and tighten import contracts to fix coupling miscounts (#474)
• Expose WeightedViolations so architecture ComplianceScore is reproducible (#476)

Fixed

• Populate SystemAnalysisResponse.Summary so analyze reports aggregate system metrics correctly (#472)

Documentation

• Add French translation (#470)
• Tidy mkdocs header logo (#469)

Fixed

• Preserve complexity metric fallbacks and use AST counts for consistent reporting (#466)
• Compute module-level cognitive complexity metrics (#464)
• Honor APTED cost model and tighten distance handling for large trees (#454)

Documentation

• Clarify that large-tree APTED is a bounded heuristic for clone detection (#467)

Testing

• Add LCOM use case unit tests (#459)
• Add unit tests for app/di_antipattern_usecase.go (#461)

Fixed

• Gate Type-4 clone classification on minimum CFG cyclomatic complexity to cut linear-shape false positives (#450)

Performance

• Speed up the full analyze pipeline by reusing parsed files and the system dependency graph, scoring APTED pairs once, and using integer LSH candidates (#438)

Fixed

• Count rule invocations (not violations) in architecture ComplianceScore denominator (#448, #426)
• Exit non-zero when output generation fails in the analyze command (#437, #249)

Added

• Honor semantic Type-4 clone classification with unified parser traversal and DFA field contracts (#435)

Fixed

• Drop strict-subset members from clone groups and keep dedupe outputs consistent (#432, #416)
• Auto-enable LSH for high pair counts in clone detection (#433, #427)

Fixed

• Fix DFA extraction for with statement alias targets (#424)
• Honor analyze config flags by preserving system defaults and validating selection (#425)
• Avoid resolving stdlib imports to shadowing project modules (#429)
• Structurally model with-statement aliases in DFA via NodeWithItem (#430)

Fixed

• Report module-scope code as <module> instead of __main__ in complexity output (#417)
• Include with context expressions in LCOM4 instance variable access detection (#419)

Fixed

• Tighten Type-2 clone classification to reduce class body false positives (#411)
• Preserve f-string interpolation expressions so LCOM4 detects self.<ivar> accesses (#415)

Added

• Main sequence architecture metrics with abstractness, instability, and zone classification (#407)

Documentation

• Sync ja/zh schemas with AbstractClassCount field (#409)

Fixed

• Emit real line range for synthetic __main__ complexity record (#403)

Documentation

• Sync rule catalog and config reference with v1.13–v1.18 releases (#406)

Added

• Package responsibility analysis with cohesion and SRP checks (#398)

Fixed

• Require text match for Type-1 clones and cap non-textual similarity (#399)
• Make clone trees expression-aware to avoid mismatched payloads (#401)

Performance

• Optimize complete-linkage grouping with sparse matrix and cached similarities (#391)

Fixed

• Make grouping clone-type tie breaks deterministic (#389)

Documentation

• Add user-facing documentation site with MkDocs Material (#387)
• Default to uvx and link README to docs site (#388)
• Add Japanese and Simplified Chinese translations (#392)

Testing

• Add clone detector benchmarks (#393)

CI

• Auto-assign PR author on open/reopen (#390)

Added

• DI (Dependency Injection) anti-pattern detection for Python code (#308)

Fixed

• Honor complexity config settings in analyze (#382)
• Preserve clone show_content config in analyze (#384)

Refactored

• Remove internal/config import from AnalyzeUseCase (#383)
• Reuse visited state in return reachability walk (#385)

Added

• Add raw code metrics to complexity analysis (#380)

Refactored

• Use system analysis in check command (#379)

Added

• Add Cognitive Complexity metric with SonarQube-style calculation (#370)
• Improve display of __main__ top-level code suggestions (#375)

Refactored

• Remove top suggestions from terminal output (#369)
• Refactor AnalyzeUseCase to depend on AnalyzeOutputFormatter interface (#376)

Added

• Add neutral_prefixes config for architecture layer detection (#365)

Fixed

• Add missing IsAbstract and Description fields to TOML layer/rule types (#362)
• Use match position as tie-breaker in findLayerForModule (#361)
• Replace detectLayerFromModule heuristic with findLayerForModule (#360)
• Resolve layer misclassification for ambiguous module paths (#359)
• Propagate architecture layers and rules from TOML config (#356)

Refactored

• Extract color hex values into CSS custom properties and Go constants (#351)

Testing

• Add architecture layer integration tests (#363)

Added

• Add actionable suggestions to analyze output (#347)

Dependencies

• Bump the go-dependencies group with 2 updates (#349)

Fixed

• Generate default_config.toml dynamically from domain defaults (#343)

Style

• Restyle HTML output with professional data-first palette (#345)

Documentation

• Update README.md

Added

• Add LCOM4 class cohesion metric (#340)
• Show README badge snippet after health score output (#331)

Fixed

• Shorten HTML tab labels to prevent wrapping (#341)
• Exclude Type-3 from default enabled clone types (#338)
• Correct APTED key-root DP ordering (#337)

Performance

• Add Jaccard pre-filter to minimize APTED calls (#339)

Documentation

• Add algorithm documentation for all analysis engines (#330)
• Redesign README header with centered logo (#329)

Testing

• Add unit tests for MCP server handlers (#332)

Fixed

• Calibrate clone detection thresholds and scoring (#327)
• Improve CBO analysis with Union type support and score calibration (#326)
• Extend clone memory test timeout to 180s (#325)

Added

• Add CFG support for Walrus operator (#322)
• Add missing error code constants (#320)

Fixed

• Improve config file discovery from target paths (#323)

Enhanced

• Add test coverage for ast_features.go (#321)