Documents the key shifts: compact-first outputs, pipeline slimming,
and DB staleness awareness. Explains why token economy is the priority.

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

- Add OutputFormat enum and OutputOptions shared helper
- Add docs/output-contract.md specifying compact format rules
- Update agent-facing commands to default to compact format:
  - hotspots, dead-code, coupling, callgraph, impact, context
- Compact format: one line per item, bounded lists, stable IDs
- JSON schema uses consistent field names (methodId, items, metadata)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

- Add MethodResolver helper for consistent method resolution
- Add --id option to context, callgraph, impact, similar commands
- Resolution precedence: --id > exact match > substring match
- Context command now prints method ID for agent copy-paste
- Update output-contract.md and LLM-QUICKSTART.md with --id examples

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

- Save analysis metadata: analyzed_at, solution_path, tool_version, git_commit
- Add 'status' command to show db info and staleness check
- Staleness heuristics: git commit change, source file modification times
- Add GitHelpers.GetCurrentCommitHash() and GetLastModifiedTime()
- Compact, table, and JSON output formats supported

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

- Add --stages option to analyze command (default: core)
- core: all stages except clustering (fast, essential features)
- full: all stages including intent clustering
- Save stages metadata for status command
- ClustersCommand shows helpful message when run after core analysis

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

- Update LLM-QUICKSTART.md with new features (status, --stages, --id)
- Trim README from 329 to 109 lines (67% reduction)
- Add links to detailed docs (output-contract, LLM quickstart)
- Focus README on essentials: install, quick start, command table

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

- ContextHandler: Include MethodId in output for agent copy-paste
- QueryHandler hotspots: Compact one-line-per-item format
- QueryHandler dead-code: Add top parameter, compact format
- All handlers now default to bounded outputs

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

- Add SnapshotTests.cs with golden file comparisons for:
  - hotspots (compact/json)
  - dead-code (compact/json)
  - callgraph (compact/json)
  - impact (compact/json)
  - coupling (compact/json)
  - context (compact/json)
  - tree (compact)

- Add snapshot update workflow: UPDATE_SNAPSHOTS=1 dotnet test
- Add docs/snapshot-testing.md documenting the workflow
- Fix CliCommandTests JSON key expectation (hotspots -> items)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Implements Task 74: Graph Traversal Engine with Configurable Strategies

Core components:
- TraversalTypes.cs: Direction, Strategy, Ranking enums
- TraversalConfig.cs: Configuration record with validation
- TraversalResult.cs: Node, Edge, and Result records
- FilterConfig.cs: Namespace/type/accessibility filtering
- GraphTraversalEngine.cs: BFS/DFS traversal with ranking

Features:
- BFS and DFS traversal strategies
- Direction control: Callers, Callees, Both
- Configurable depth limits and max results
- Four ranking strategies:
  - BlastRadius: Transitive caller count
  - Complexity: Cognitive complexity from metrics
  - Coupling: Afferent + Efferent coupling
  - Combined: Weighted normalized combination
- Session-level caching for performance
- Filter support for namespaces, types, accessibility

Tests: 37 new tests covering types, traversal, and ranking

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Implements Task 76.1: Core layer detection infrastructure

- ArchitecturalLayer enum: Presentation, Application, Domain,
  Infrastructure, Shared, Unknown
- LayerAssignment record for storing detection results
- LayerDetector with pattern-based detection:
  - Default patterns for Clean Architecture/DDD
  - Type name hints (Controller, Repository, Handler suffixes)
  - Confidence scoring (1.0 for exact match, 0.5 for partial)
- Dependency validation rules for Clean Architecture
- 12 unit tests covering detection and validation

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Implements Task 76.2: Storage methods for layer assignments

Schema changes:
- Add TypeLayers table (TypeId, Layer, Confidence, Reason)
- Add IX_TypeLayers_Layer index

IStorageService interface additions:
- SaveLayerAssignmentsAsync
- GetLayerAssignmentsAsync
- GetLayerForTypeAsync

StorageService implementation with proper transaction handling.

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Implements RefineByDependencyDirectionAsync() that analyzes call graph
dependencies between types and adjusts layer assignment confidence based
on Clean Architecture dependency rules:

- Lowers confidence when types violate allowed dependency directions
  (e.g., Domain depending on Infrastructure)
- Boosts confidence for low-confidence types with consistent valid deps
- Clamps confidence to minimum 0.1 to prevent negative values
- Adds violation warnings to Reason field

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Implements LayersCommand following ICommandHandler pattern:
- Displays type-to-layer assignments with confidence scores
- Supports filtering by layer (--layer) and minimum confidence
- Outputs in compact, table, JSON, or CSV formats
- Shows violations with [!] marker in compact mode

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Implements blast radius computation using BFS on reverse call graph:
- Counts direct and transitive callers for each method
- Computes depth (max distance from entry points)
- Identifies entry points (methods with no callers) that can trigger code
- Performance: handles 1000+ methods in <2 seconds

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

- Extends schema with BlastRadius and BlastDepth columns
- Adds index on BlastRadius for efficient sorting
- Implements SaveBlastRadiusAsync using UPSERT pattern
- Updates GetMethodMetricsAsync to return blast radius data

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

- Adds ComputeBlastRadiusStage to AnalysisStageHelpers
- Runs after StoreResultsStage when call graph is available
- Shows verbose output with max blast radius and high-impact count
- Persists results to Metrics table via SaveBlastRadiusAsync

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Hotspots command:
- Adds --sort option (complexity|blast-radius|risk)
- Shows blast radius in output when sorting by blast or risk
- Includes risk score: complexity * log(blast_radius + 1)

Context command:
- Shows blast radius with depth and computed risk score

Updates GetHotspotsWithThresholdAsync to include BlastRadius/BlastDepth
and support custom sort ordering.

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Defines complete query model with:
- QuerySeed: starting points (MethodId, Pattern, Namespace, Cluster)
- QueryExpand: traversal control (Direction, MaxDepth, Transitive)
- QueryFilter: inclusion/exclusion rules (namespaces, types, complexity)
- QueryRank: result ordering (BlastRadius, Complexity, Coupling, Combined)
- QueryOutput: format and limits (Compact/Json/Table, MaxResults)

Supporting enums: ExpandDirection, RankStrategy, QueryOutputFormat

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Implements validation rules:
- Seed must have at least one non-null property
- MaxDepth bounds (0-100)
- MinComplexity <= MaxComplexity
- No overlapping Include/Exclude namespaces
- MaxResults bounds (1-1000)
- Empty/whitespace checks for all string properties

Includes ValidationResult record and extension method for fluent usage.

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Implements GraphQueryExecutor that:
- Validates GraphQuery via GraphQueryValidator before execution
- Resolves seeds (MethodId, MethodPattern, Namespace, Cluster)
- Translates GraphQuery to TraversalConfig for GraphTraversalEngine
- Handles ExpandDirection.None to return seed-only results
- Applies ranking strategies (Complexity, BlastRadius, Coupling, Combined)
- Formats results with optional metrics and location info
- Supports MaxResults limiting and execution time tracking

Includes 19 comprehensive tests covering all query scenarios.

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Implements QueryPlanCache with:
- Thread-safe ConcurrentDictionary storage
- LRU eviction when cache exceeds max size (default 100)
- Time-based expiration (default 5 minutes)
- SHA256-based query hashing (excludes Output settings)
- Hit/miss tracking with GetStats() method

GraphQueryExecutor integration:
- Optional useCache parameter (default true)
- ClearCache() and GetCacheStats() methods
- Caches resolved seeds and expand direction

Includes 13 QueryPlanCache tests and 4 caching integration tests.

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

GraphQuerySerializer provides:
- JSON serialization with camelCase naming and enum strings
- Deserialization with TryDeserialize error handling
- JSON Schema generation (draft-07) with full property docs

QueryCommand CLI command:
- Execute queries from --query-file or inline JSON argument
- --schema flag outputs JSON schema
- Supports all output formats (compact, json, table)
- Validates queries before execution

Includes 13 serializer tests covering round-trip, enums, and schema.

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

DependencyRuleEngine provides:
- Glob pattern matching for source/target namespaces
- Built-in Clean Architecture rules (12 default rules)
- Custom rules via JSON file with includeDefaults option
- Violation detection with file:line locations
- Severity levels (Error, Warning, Info)

check-deps CLI command:
- --rules <file> for custom rules
- --show-rules to display loaded rules
- --sample to generate example rules.json
- Groups violations by rule in output
- JSON output format supported

Includes 18 tests covering pattern matching, rule loading, and detection.

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

- Added --seed option for quick method pattern or ID matching
- Added --depth, --direction, --rank, --top options for common parameters
- Kept --json and --query-file for full JSON query support
- Auto-detects pattern vs exact ID based on wildcards (* or ?)
- Updated command description for agent usage

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

- Exposes GraphQueryExecutor via MCP as cg_query tool
- Supports seed pattern, direction, depth, rank, and top parameters
- Auto-detects wildcards to use pattern vs exact ID lookup
- Excludes test methods by default
- Token-optimized compact response format

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

- ProtectedZone model with DoNotModify, RequireApproval, Deprecated levels
- JSON config loading from .ai-code-graph/protected-zones.json
- Glob pattern matching for method/namespace/type identification
- Methods for checking protection and filtering protected methods
- 20 unit tests covering all functionality

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

…d MCP

- Context command: shows warning if method is in protected zone
- Impact command: lists protected methods in blast radius
- Callgraph command: marks protected methods in call graph
- MCP cg_query: includes protection warnings in results

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

- Updated CLI help text for search commands to point to query
- Updated MCP tool descriptions to indicate search is fallback
- Updated slash commands with deprecation notes
- Updated CLAUDE.md with recommended workflow (query first)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

All tasks completed in this session:
- Task 80: Graph-First Query CLI Command
- Task 81: MCP Graph Query Tool
- Task 79: Protected Zone Marking
- Task 83: Architectural Summary in Context
- Task 82: Deprecate Token Search
- Task 71: Benchmark artifacts gitignore (was already done)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

- Add missing slash commands: query, status, layers, check-deps
- Update CLAUDE.md with all 21 user-facing commands
- Update SetupClaudeCommand.cs to generate all command files
- Add content generators for: impact, dead-code, coupling, diff,
  semantic-search, query, status, layers, check-deps

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>