* chore: ignore .worktrees/ for subagent-driven development (#7) Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * chore: add Roslyn, OpenAI, sqlite-vec dependencies (#7) * refactor: remove redundant Microsoft.Build.* overrides (#7) Code review (I-1) found that Microsoft.Build, Microsoft.Build.Framework, and Microsoft.Build.Utilities.Core overrides at 17.14.28 were redundant — only Microsoft.Build.Tasks.Core needs the explicit pin to silence NU1903. Removed the three unnecessary entries. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat(core): add CodeChunk record and supporting types (#7) Introduces pure-data records and constant classes for the codebase indexer: CodeChunk, ChunkModifiers, ChunkAttribute, ChunkParameter, ChunkGenericTypeParameter, plus SymbolKinds and Accessibilities. Adds FluentAssertions to the Unit test project to support shape-lock tests, and tightens the async-suffix architecture rule to skip property accessors (synthesised get_/set_ methods). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat(core): add ISourceTextHashingService (#7) Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat(core): SqliteIndexStore opens DB and round-trips metadata (#7) Adds the v1 SQL schema as an embedded resource, an internal IIndexStore abstraction, and a SqliteIndexStore that opens the file, loads the sqlite-vec extension, applies the schema on first run, and round-trips the index_metadata row. Reopening with a stale schema_version throws. Tests.Unit is added to the Directory.Build.targets vec0 copy condition so the open-the-file unit tests can find the native binary. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat(core): IndexStore writes chunks, child rows, and embeddings (#7) Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat(core): IndexStore reads summaries and supports targeted deletion (#7) Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat(core): ChunkExtractor extracts type-level chunks (#7) Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat(core): ChunkExtractor extracts methods, ctors, operators, indexers (#7) Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat(core): ChunkExtractor extracts properties, fields, events (#7) Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat(core): ChunkExtractor handles attributes and partial types (#7) Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat(core): ChunkExtractor skips generated and out-of-repo files (#7) Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat(core): pure-function ReconciliationService with insert/update/delete plan (#7) Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat(core): MsBuildWorkspaceLoadingService opens .sln/.slnx files (#7) Adds IWorkspaceLoadingService and MsBuildWorkspaceLoadingService for loading Roslyn solutions via MSBuildWorkspace. Disposes the prior workspace on each Open call to avoid leaks across reuse. Bumps Microsoft.CodeAnalysis.{CSharp.Workspaces,Workspaces.MSBuild} from 4.12.0 to 5.0.0 because 4.12 routes .slnx through the legacy SolutionFile parser, which fails with "No file format header found". Adds OpenSolutionAsync to the framework Async-method whitelist in NamingConventionTests so the interface can mirror MSBuildWorkspace's own API. Adds FluentAssertions to the Integration project. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * feat(core): CliGitDiffService exposes head sha and changed files (#7) * feat(core): OpenAIEmbeddingClient with batching and retry (#7) * feat(core): IndexingService orchestrates full and incremental runs (#7) * feat(cli): add 'index' verb (#7) * test(integration): sample fixture solution and indexer fakes (#7) Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * test(integration): cover indexer scenarios end-to-end (#7) Adds 8 integration scenarios for IndexingService covering cold index, no-op rerun, single-method edit, new-class insert, file deletion, csproj reparse, slnx-triggered full reindex, and a filtered KNN-style query against the populated SQLite store. Adds Reset() to FakeEmbeddingClient and makes UserService.FindAsync genuinely async to exercise the is_async metadata column. Two scenarios fail and reveal pre-existing indexer defects: - Should_remove_all_chunks_when_a_file_is_deleted: ProcessIncrementalFile bypasses the plan's Deletes accumulator when a file is missing, so IndexRunResult.DeletedChunks is always 0 even though the rows are gone. - Should_reparse_whole_project_when_csproj_changes: ChunkExtractor's ToFullyQualifiedName returns the bare member name for symbols (e.g. ILogger.Log and ConsoleLogger.Log both produce "Log"), so ReconciliationService.Plan throws on the duplicate dictionary key. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * fix(core): correct DeletedChunks counter and FQN collisions (#7) Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com> * fix(core): array attributes and idempotent schema apply (#7) Surfaced by smoke-testing against a real net48 codebase: - ChunkExtractor.BuildAttributes crashed on attributes with array arguments (e.g. params object[]) because TypedConstant.Value throws when Kind == Array. Added recursive CoerceTypedConstant helper that uses .Values for arrays. - SqliteIndexStore.ApplySchemaAsync failed re-opening a partially written index from a crashed prior run. Schema DDL now uses CREATE TABLE/INDEX/VIRTUAL TABLE IF NOT EXISTS so re-open is safe. --------- Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Implement codebase indexer per spec #7

Description

Metadata

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

Issue actions