flutter_ai_context 0.7.0

A local-first project intelligence layer for Flutter codebases. Give AI a real understanding of your Flutter project.

Changelog

0.7.0

Added

• MCP server (Phase 1): flutter_ai_context mcp over stdio
• MCP tools: project_status, list_features, get_context
• MCP resources: flutter-ai-context://agents, architecture, context/{scope}

0.6.0

Added

• Graph-based hybrid state management detection (architectural nodes outweigh Consumer/context.read signals)
• Frozen graph schema v1 contract: doc/GRAPH_SCHEMA_v1.md
• Cursor onboarding rule: .cursor/rules/flutter-ai-context.mdc on init (context.generate_cursor_rule)
• sync + status FRESH/STALE as the documented default workflow

Improved

• sync and status report context freshness, changed file counts, and schema version
• init prints next-step workflow (sync after edits, scan for full rebuilds)
• Schema mismatch on cache load triggers rebuild with a clear message
• Graph-first primary SM with DI-layer detection (GetX architecture + Provider widget signals)
• Human-readable docs: doc/architecture.md, doc/GRAPH_SCHEMA_v1.md

0.5.3

Improved

• Shared flow inference distinguishes GetX, Riverpod, Provider, and StatefulWidget patterns
• Context packs include Agent Guide, Entry Points, and a dedicated Widgets section
• context.read<T>() / context.watch<T>() edges improve repository and controller detection
• Faster scans via AST-based semantic skip for constants, theme, and non-interactive files

0.5.2

Improved

• Scan performance: parse-once AST cache, selective semantic resolution, batched resolves
• Context ranker includes core/ and shared/ files imported directly by a feature
• Context ranker supports modules/ feature paths
• GetX hybrid detection for ChangeNotifier controllers and *Binding classes
• Context packs dedupe Related Nodes and use incremental token budgeting

0.5.1

Fixed

• Detect *Page / pages/ widgets as screens (GetX and page-first projects)
• Context ranker scopes results to the target feature folder (less cross-feature noise)
• Context pack signatures prefer the primary class per file over inner types
• Doctor separates warnings vs info; feature.missing_screen respects Page widgets
• Doctor readiness score ignores info-severity findings
• context CLI lists only files included in the token-budgeted pack

Added

• getx_pages benchmark fixture (24 Dart files, page-first GetX layout)
• Tests for doctor scoring and context pack selection

0.5.0

Added

• flutter_ai_context context --list to discover available feature scopes
• Class signatures in context packs (supertype + key methods from graph metadata)
• Import-graph sync invalidation (re-scan files that import changed modules)
• Lazy per-file semantic resolution for faster scans
• Golden test for attendance context pack output

Improved

• Doctor screen_bypasses_state_management ignores test/generated files and watch/read edges
• Context pack file entries now include compact class signatures

0.4.0

Added

• Graph-distance context pack ranking with role-aware scoring (screen → state → data)
• Context packs grouped by layer with observed feature flow
• Configurable context token budget (context.token_budget, default 2000)
• Scope suggestions when context <scope> does not match a feature
• Transitive sync invalidation across dependent graph nodes

Changed

• context output now includes score, observed flow, and grouped sections

0.3.0

Added

• Expanded doctor rules: declared vs observed state management, http in UI, screen bypassing state layer, feature missing screen, routing mismatch, provider suffix
• auto_route detection (AutoRoute, @RoutePage, @AutoRoute)
• Cross-file sync invalidation for direct graph dependents
• Element2 fragment fallback for supertype resolution
• Benchmark tests for fixture scan performance
• Fixture: auto_route_feature

Fixed

• status now reports total indexed files instead of 0 after incremental sync
• features.md deduplicates file paths per feature cluster

0.2.1

• Re-publish under verified publisher sangga.id (transfer via pub.dev Admin)

0.2.0

Added

• Semantic session via AnalysisContextCollection with AST fallback
• Symbol registry for stable graph relationship targets
• GoRouter / Navigator route detection and .ai/routes.md generation
• Riverpod/GetX signal detector (ref.watch, @riverpod, GetxController, Obx, .obs)
• Golden tests for generated Markdown
• Fixtures: Riverpod, Bloc clean, GetX, migration scenario
• Release notes: doc/RELEASE_v0.2.0.md

Fixed

• flutter_ai_context.yaml writer now quotes glob patterns ('**/*.g.dart') to avoid YAML alias errors

Dogfood

• Validated on lms_intelecto (209 Dart files, 10 features, Provider + Dio)

0.1.0

• Initial release
• CLI commands: init, scan, sync, status, doctor, context
• Analyzer-backed Dart scanning and Project Knowledge Graph
• Architecture/state management inference with confidence/evidence
• Feature clustering (feature-first and layer-first)
• Context generation (.ai/*.md, AGENTS.md)
• Incremental cache with file hashing
• Doctor rule engine with AI readiness score
• Focused context packs with token estimation
• Test fixtures and CI pipeline