Codebase intelligence for AI and humans: code health scores, auto-generated docs, git analytics, dead code detection, and architectural decisions via MCP.
Python aicode-complexitycode-healthcode-intelligencecode-qualitydead-codedeveloper-toolsdocumentationgit-analyticsmcpopen-sourcepythonrefactoringstatic-analysistechnical-debt
fix(answer): never silently drop synthesis when a usable API key exists (#888)
Swati Ahujaf2fb28f443b0
diff --git a/packages/server/src/repowise/server/mcp_server/tool_answer/answer.py b/packages/server/src/repowise/server/mcp_server/tool_answer/answer.pyindex b6aac06e6..e726592b7 100644--- a/packages/server/src/repowise/server/mcp_server/tool_answer/answer.py+++ b/packages/server/src/repowise/server/mcp_server/tool_answer/answer.py@@ -923,15 +923,23 @@ async def get_answer(provider = _resolve_provider_for_answer(getattr(ctx, "path", None))if provider is None:# Retrieval-only mode (no provider). Return the hits so the agent can- # at least skip the search_codebase step.- return {+ # at least skip the search_codebase step — but mark the degradation+ # loudly: an arm/user should never need to diff payload shapes to+ # notice synthesis is unplugged.+ _log.warning(+ "get_answer running WITHOUT synthesis: no LLM provider resolvable "+ "(set REPOWISE_PROVIDER + its API key, or any supported API key)."+ )+ payload = {"answer": "","citations": [],"confidence": "low",+ "degraded": "no-llm-provider","fallback_targets": fallback_targets,"retrieval": _serialize_hits(hits),"note": (- "No LLM provider configured (set REPOWISE_PROVIDER + API key). "+ "DEGRADED: no LLM provider configured (set REPOWISE_PROVIDER "+ "+ API key). ""Returning retrieval hits only — Read the listed files to answer."),"_meta": _build_meta(@@ -941,6 +949,8 @@ async def get_answer(targets=fallback_targets,),}+ payload["_meta"]["degraded"] = "no-llm-provider"+ return payload# Decision fusion (why-shaped questions only) + structured prelude. Both# layers are gated on signal: no ADRs for the top hits → no decisions@@ -979,9 +989,10 @@ async def get_answer("answer": "","citations": [],"confidence": "low",+ "degraded": "synthesis-failed","fallback_targets": fallback_targets,"retrieval": _serialize_hits(hits),- "note": f"LLM synthesis failed ({type(exc).__name__}). Read the listed files to answer.",+ "note": f"DEGRADED: LLM synthesis failed ({type(exc).__name__}). Read the listed files to answer.","_meta": _build_meta(timing_ms=(time.perf_counter() - t0) * 1000,hint=_answer_hint("low", len(hits)),diff --git a/packages/server/src/repowise/server/mcp_server/tool_answer/synthesis.py b/packages/server/src/repowise/server/mcp_server/tool_answer/synthesis.pyindex 767a96e40..941c0db2d 100644--- a/packages/server/src/repowise/server/mcp_server/tool_answer/synthesis.py+++ b/packages/server/src/repowise/server/mcp_server/tool_answer/synthesis.py@@ -42,6 +42,7 @@ def _load_repo_provider_config(if repo_path is None:return None, None, {}+ config_path = repo_path / ".repowise" / "config.yaml"state_path = repo_path / ".repowise" / "state.json"env_path = repo_path / ".repowise" / ".env"@@ -49,11 +50,27 @@ def _load_repo_provider_config(model: str | None = Noneoverlay: dict[str, str] = {}+ # config.yaml first: it is the user-editable intent. state.json only+ # records what the LAST index run used, which goes stale the moment the+ # user switches providers (observed: a config saying openai while+ # state.json still said gemini from a months-old index build — resolution+ # followed state.json into a keyless provider and synthesis went dark).+ try:+ if config_path.is_file():+ import yaml++ data = yaml.safe_load(config_path.read_text(encoding="utf-8")) or {}+ if isinstance(data, dict):+ name = data.get("provider") or None+ model = data.get("model") or None+ except Exception:+ _log.debug("Failed to read %s", config_path, exc_info=True)+try:if state_path.is_file():data = _json.loads(state_path.read_text(encoding="utf-8"))- name = data.get("provider") or None- model = data.get("model") or None+ name = name or data.get("provider") or None+ model = model or data.get("model") or Noneexcept Exception:_log.debug("Failed to read %s", state_path, exc_info=True)@@ -126,25 +143,42 @@ def _resolve_provider_for_answer(repo_path: Path | None = None):return valreturn None- # Explicit selection wins.+ # Explicit selection wins — when its key is actually available. A+ # configured/persisted provider whose key is absent must NOT end+ # resolution: returning None here made get_answer silently drop to+ # retrieval-only mode even though another provider's key sat in the env+ # (the auto-detect below was unreachable). Fall through instead, and+ # drop the persisted model on the way down — it belongs to the named+ # provider and would break a cross-provider fallback call.+ _KEY_ENVS = {+ "anthropic": ("ANTHROPIC_API_KEY",),+ "openai": ("OPENAI_API_KEY",),+ "deepseek": ("DEEPSEEK_API_KEY",),+ "kimi": ("KIMI_API_KEY",),+ "gemini": ("GEMINI_API_KEY", "GOOGLE_API_KEY"),+ }if name:- kw: dict[str, Any] = {}- if model:- kw["model"] = model- if name == "anthropic" and _env("ANTHROPIC_API_KEY"):- kw["api_key"] = _env("ANTHROPIC_API_KEY")- elif name == "openai" and _env("OPENAI_API_KEY"):- kw["api_key"] = _env("OPENAI_API_KEY")- elif name == "deepseek" and _env("DEEPSEEK_API_KEY"):- kw["api_key"] = _env("DEEPSEEK_API_KEY")- elif name == "kimi" and _env("KIMI_API_KEY"):- kw["api_key"] = _env("KIMI_API_KEY")- elif name == "gemini" and (_env("GEMINI_API_KEY") or _env("GOOGLE_API_KEY")):- kw["api_key"] = _env("GEMINI_API_KEY") or _env("GOOGLE_API_KEY")- base_url = _resolve_base_url(name)- if base_url:- kw["base_url"] = base_url- return _try(name, **kw)+ key_envs = _KEY_ENVS.get(name)+ key = next((v for v in map(_env, key_envs or ()) if v), None)+ if key_envs is None or key:+ kw: dict[str, Any] = {}+ if model:+ kw["model"] = model+ if key:+ kw["api_key"] = key+ base_url = _resolve_base_url(name)+ if base_url:+ kw["base_url"] = base_url+ provider = _try(name, **kw)+ if provider is not None:+ return provider+ _log.warning(+ "Configured provider %r has no usable key/setup; falling back to "+ "auto-detection from available API keys.",+ name,+ )+ if not (os.environ.get("REPOWISE_DOC_MODEL") or os.environ.get("REPOWISE_MODEL")):+ model = None # persisted model was provider-specific# Auto-detect from API keys.if _env("ANTHROPIC_API_KEY"):diff --git a/tests/unit/server/mcp/test_provider_resolution.py b/tests/unit/server/mcp/test_provider_resolution.pynew file mode 100644index 000000000..68ce68e9e--- /dev/null+++ b/tests/unit/server/mcp/test_provider_resolution.py@@ -0,0 +1,85 @@+"""get_answer provider resolution: intent precedence and keyless fall-through.++Regression coverage for the silent retrieval-only degradation: an index+whose state.json recorded a provider from a months-old build, in an env+that only carried a different provider's key, resolved to None and ran+get_answer without synthesis — no error, no fallback, even though a usable+key was sitting in the environment.+"""++from __future__ import annotations++import json as _json++import pytest++from repowise.server.mcp_server.tool_answer.synthesis import (+ _load_repo_provider_config,+ _resolve_provider_for_answer,+)++[email protected](autouse=True)+def _clean_env(monkeypatch):+ for var in ("REPOWISE_PROVIDER", "REPOWISE_DOC_MODEL", "REPOWISE_MODEL",+ "OPENAI_API_KEY", "ANTHROPIC_API_KEY", "GEMINI_API_KEY",+ "GOOGLE_API_KEY", "DEEPSEEK_API_KEY", "KIMI_API_KEY",+ "OLLAMA_BASE_URL", "OPENAI_BASE_URL"):+ monkeypatch.delenv(var, raising=False)+++def _write_repo(tmp_path, *, config=None, state=None):+ d = tmp_path / ".repowise"+ d.mkdir(parents=True, exist_ok=True)+ if config is not None:+ (d / "config.yaml").write_text(+ "\n".join(f"{k}: {v}" for k, v in config.items()), encoding="utf-8")+ if state is not None:+ (d / "state.json").write_text(_json.dumps(state), encoding="utf-8")+ return tmp_path+++def test_config_yaml_outranks_state_json(tmp_path):+ """config.yaml is user intent; state.json only records the last run."""+ repo = _write_repo(+ tmp_path,+ config={"provider": "openai", "model": "gpt-5.4-nano"},+ state={"provider": "gemini", "model": "gemini-3.1-flash-lite-preview"},+ )+ name, model, _ = _load_repo_provider_config(repo)+ assert name == "openai"+ assert model == "gpt-5.4-nano"+++def test_state_json_still_used_when_config_lacks_provider(tmp_path):+ repo = _write_repo(tmp_path, config={"commit_limit": 200},+ state={"provider": "openai", "model": "gpt-5.4-nano"})+ name, model, _ = _load_repo_provider_config(repo)+ assert name == "openai"+ assert model == "gpt-5.4-nano"+++def test_keyless_persisted_provider_falls_back_to_available_key(tmp_path, monkeypatch):+ """gemini persisted, only OPENAI_API_KEY present -> openai provider, not None."""+ repo = _write_repo(tmp_path,+ state={"provider": "gemini", "model": "gemini-3.1-flash"})+ monkeypatch.setenv("OPENAI_API_KEY", "sk-test")+ provider = _resolve_provider_for_answer(repo)+ assert provider is not None+ assert getattr(provider, "provider_name", "") == "openai"+ # The persisted gemini model must not leak into the openai call.+ assert "gemini" not in (getattr(provider, "model_name", "") or "")+++def test_no_keys_at_all_resolves_to_none(tmp_path):+ repo = _write_repo(tmp_path, state={"provider": "gemini"})+ assert _resolve_provider_for_answer(repo) is None+++def test_explicit_env_provider_still_wins(tmp_path, monkeypatch):+ repo = _write_repo(tmp_path, state={"provider": "gemini"})+ monkeypatch.setenv("REPOWISE_PROVIDER", "openai")+ monkeypatch.setenv("OPENAI_API_KEY", "sk-test")+ provider = _resolve_provider_for_answer(repo)+ assert provider is not None+ assert getattr(provider, "provider_name", "") == "openai"