|
|
ec55984fd8
|
feat: Phase 5 polish — conditional LSP tools, capabilities, sidecar hooks
- ToolSide enum: documented Sidecar future variant
- StartSession.capabilities: client reports LSP availability
- Client detects LSP binaries on PATH, sends ["lsp_rust", "lsp_typescript"]
- build_tool_definitions() conditionally registers LSP tools only when
client has LSP capability — model won't hallucinate unavailable tools
- CodeSession stores capabilities, has_lsp(), has_capability() accessors
- git_branch() reads from git for breadcrumb scoping
- ToolRegistry.gitea_client() accessor for reindex endpoint
|
2026-03-24 09:54:14 +00:00 |
|
|
|
c213d74620
|
feat: code search tool + breadcrumb context injection + integration tests
search_code tool:
- Server-side tool querying sol_code OpenSearch index
- BM25 search across symbol_name, signature, docstring, content
- Branch-aware with boost for current branch, mainline fallback
- Registered in ToolRegistry execute dispatch
Breadcrumb injection:
- build_context_header() now async, injects adaptive breadcrumbs
- Hybrid search: _analyze → wildcard symbol matching → BM25
- Token budget enforcement (default outline + relevant expansion)
- Graceful degradation when OpenSearch unavailable
GrpcState:
- Added Option<OpenSearch> for breadcrumb retrieval
- code_index_name() accessor
Integration tests (6 new, 226 total):
- Index + search: bulk index symbols, verify BM25 retrieval
- Breadcrumb outline: aggregation query returns project structure
- Breadcrumb expansion: substantive query triggers relevant symbols
- Token budget: respects character limit
- Branch scoping: feat/code symbols preferred over mainline
- Branch deletion: cleanup removes branch symbols, mainline survives
|
2026-03-24 00:19:17 +00:00 |
|
|
|
40a6772f99
|
feat: 13 e2e integration tests against real Mistral API
Orchestrator tests:
- Simple chat roundtrip with token usage verification
- Event ordering (Started → Thinking → Done)
- Metadata pass-through (opaque bag appears in Started event)
- Token usage accuracy (longer prompts → more tokens)
- Conversation continuity (multi-turn recall)
- Client-side tool dispatch + mock result submission
- Failed tool result handling (is_error: true)
- Server-side tool execution (search_web via conversation)
gRPC tests:
- Full roundtrip (StartSession → UserInput → Status → TextDone)
- Client tool relay (ToolCall → ToolResult through gRPC stream)
- Token counts in TextDone (non-zero verification)
- Session resume (same room_id, resumed flag)
- Clean disconnect (EndSession → SessionEnd)
Infrastructure:
- ToolRegistry::new_minimal() — no OpenSearch/Matrix needed
- ToolRegistry fields now Option for testability
- GrpcState.matrix now Option
- grpc_bridge moved to src/grpc/bridge.rs
- TestHarness loads API key from .env
|
2026-03-23 20:54:28 +00:00 |
|
