Fix ruff format & CI test

Author: Winda0001

.github/workflows/ci.yml

@@ -59,8 +59,10 @@ jobs:
         # Exit code 5 = "no tests collected" — treat as pass until tests are added.
         # Real failures (exit code 1) still fail CI.
         run: |
+          set +e
           pytest tests/ --tb=short -q
           STATUS=$?
+          set -e
           [ $STATUS -eq 5 ] && exit 0 || exit $STATUS
         shell: bash
 

src/agents/acquire_agent.py

@@ -19,18 +19,18 @@
 # Diagnostic Test Accuracy (DTA) filter for Diagnosis questions.
 # Targets sensitivity, specificity, ROC, likelihood ratio, and QUADAS studies.
 _DTA_FILTER = (
-    "(sensitivity[tiab] OR specificity[tiab] OR \"diagnostic accuracy\"[tiab] "
-    "OR \"likelihood ratio\"[tiab] OR \"ROC curve\"[tiab] OR \"area under the curve\"[tiab] "
-    "OR QUADAS[tiab] OR \"diagnostic test\"[tiab] OR \"predictive value\"[tiab] "
+    '(sensitivity[tiab] OR specificity[tiab] OR "diagnostic accuracy"[tiab] '
+    'OR "likelihood ratio"[tiab] OR "ROC curve"[tiab] OR "area under the curve"[tiab] '
+    'OR QUADAS[tiab] OR "diagnostic test"[tiab] OR "predictive value"[tiab] '
     "OR systematic[sb])"
 )
 
 # Observational filter for Prognosis and Harm questions.
 # Allows cohort studies; keeps systematic reviews; drops RCT bias.
 _OBSERVATIONAL_FILTER = (
     "(cohort[tiab] OR prospective[tiab] OR retrospective[tiab] "
-    "OR \"follow-up\"[tiab] OR prognosis[tiab] OR survival[tiab] "
-    "OR \"risk factor\"[tiab] OR incidence[tiab] OR mortality[tiab] "
+    'OR "follow-up"[tiab] OR prognosis[tiab] OR survival[tiab] '
+    'OR "risk factor"[tiab] OR incidence[tiab] OR mortality[tiab] '
     "OR systematic[sb])"
 )
 
@@ -86,10 +86,10 @@ def _extract_query(self, content: str) -> str:
             remainder = content[query_start:].strip()
             if "```" in remainder:
                 cb_start = remainder.find("```")
-                after_backticks = remainder[cb_start + 3:]
+                after_backticks = remainder[cb_start + 3 :]
                 newline = after_backticks.find("\n")
                 if newline >= 0:
-                    after_backticks = after_backticks[newline + 1:]
+                    after_backticks = after_backticks[newline + 1 :]
                 cb_end = after_backticks.find("```")
                 if cb_end >= 0:
                     return after_backticks[:cb_end].strip()
@@ -114,9 +114,15 @@ def _infer_study_type(self, evidence: Evidence) -> str:
         text = f"{evidence.title} {evidence.abstract or ''}".lower()
         if "systematic review" in text or "meta-analysis" in text:
             return "Systematic Review"
-        if ("randomized controlled trial" in text or "randomised controlled trial" in text or
-                "randomized clinical trial" in text or "randomised clinical trial" in text or
-                "rct" in text or " randomized " in text or " randomised " in text):
+        if (
+            "randomized controlled trial" in text
+            or "randomised controlled trial" in text
+            or "randomized clinical trial" in text
+            or "randomised clinical trial" in text
+            or "rct" in text
+            or " randomized " in text
+            or " randomised " in text
+        ):
             return "RCT"
         if "cohort study" in text or "cohort" in text:
             return "Cohort Study"
@@ -153,7 +159,9 @@ def _listwise_rank(
         lines = []
         for i, e in enumerate(candidates):
             study_hint = f"[{e.study_type}] " if e.study_type else ""
-            abstract_preview = e.key_sentences if e.key_sentences else (e.abstract or "")[:150]
+            abstract_preview = (
+                e.key_sentences if e.key_sentences else (e.abstract or "")[:150]
+            )
             lines.append(
                 f"[{i + 1}] {study_hint}{e.title}\n"
                 f"     Abstract: {abstract_preview}"
@@ -171,7 +179,9 @@ def _listwise_rank(
         )
 
         response = self.llm.invoke(prompt)
-        print(f"[DEBUG] Listwise ranking response (first 300 chars): {response.content[:300]}")
+        print(
+            f"[DEBUG] Listwise ranking response (first 300 chars): {response.content[:300]}"
+        )
 
         # Parse ranked IDs, validate, deduplicate
         try:
@@ -239,18 +249,24 @@ def execute(self, state: WorkflowState) -> Dict[str, Any]:
         try:
             t0 = time.time()
             if self._use_local_db(question_type):
-                print(f"[DEBUG] question_type={question_type}, routing to local obstetrics DB")
+                print(
+                    f"[DEBUG] question_type={question_type}, routing to local obstetrics DB"
+                )
                 raw_results = search_local(query=base_query, top_k=20)
                 search_query_used = base_query
                 print(f"[DEBUG] Local DB returned {len(raw_results)} articles")
                 print(f"[TIMING] Local DB search: {time.time()-t0:.1f}s")
             else:
                 filtered_query = self._apply_search_filter(base_query, question_type)
-                print(f"[DEBUG] question_type={question_type}, filtered query: {filtered_query}")
+                print(
+                    f"[DEBUG] question_type={question_type}, filtered query: {filtered_query}"
+                )
                 raw_results = search_pubmed(query=filtered_query, max_results=20)
                 print(f"[DEBUG] PubMed (filtered) returned {len(raw_results)} articles")
                 if len(raw_results) == 0:
-                    print("[DEBUG] Filtered query returned 0 results — falling back to base query")
+                    print(
+                        "[DEBUG] Filtered query returned 0 results — falling back to base query"
+                    )
                     raw_results = search_pubmed(query=base_query, max_results=20)
                     print(f"[DEBUG] PubMed (base) returned {len(raw_results)} articles")
                     search_query_used = base_query

src/agents/apply_agent.py

@@ -3,6 +3,7 @@
 from src.agents.base import BaseAgent, robust_parse_json
 from src.state.schema import WorkflowState, Recommendation
 
+
 class ApplyAgent(BaseAgent):
     """Agent for generating clinical recommendations"""
 
@@ -12,8 +13,10 @@ def __init__(self, llm, tools: List[Any] = None):
 
     def _load_prompt(self) -> str:
         """Load prompt template from file"""
-        prompt_path = Path(__file__).parent.parent / "config" / "prompts" / "apply_agent.txt"
-        with open(prompt_path, 'r', encoding='utf-8') as f:
+        prompt_path = (
+            Path(__file__).parent.parent / "config" / "prompts" / "apply_agent.txt"
+        )
+        with open(prompt_path, "r", encoding="utf-8") as f:
             return f.read()
 
     def _parse_json(self, content: str) -> dict:
@@ -32,16 +35,18 @@ def execute(self, state: WorkflowState) -> Dict[str, Any]:
         if state.get("backtrack_reason"):
             backtrack_context = f"Previous attempt failed: {state['backtrack_reason']}\nPlease address these issues in your recommendation."
 
-        evidence_summary = "\n\n".join([
-            f"Evidence {i+1}:\nTitle: {e.title}\nQuality: {e.grade_level}\nSource: {e.source}"
-            for i, e in enumerate(appraisal.evidence)
-        ])
+        evidence_summary = "\n\n".join(
+            [
+                f"Evidence {i+1}:\nTitle: {e.title}\nQuality: {e.grade_level}\nSource: {e.source}"
+                for i, e in enumerate(appraisal.evidence)
+            ]
+        )
 
         prompt = self.prompt_template.format(
             question=question,
             evidence_summary=evidence_summary,
             appraisal_summary=appraisal.summary,
-            backtrack_context=backtrack_context
+            backtrack_context=backtrack_context,
         )
 
         response = self.llm.invoke(prompt)
@@ -50,7 +55,9 @@ def execute(self, state: WorkflowState) -> Dict[str, Any]:
         except ValueError:
             # LLM may output reasoning only without JSON (response truncated or forgot JSON).
             # Retry once with a short follow-up prompt asking only for the JSON block.
-            print("[WARN] Apply agent: JSON parse failed on first response, retrying for JSON only.")
+            print(
+                "[WARN] Apply agent: JSON parse failed on first response, retrying for JSON only."
+            )
             retry_prompt = (
                 "Your previous response did not include a valid JSON block. "
                 "Please output ONLY the following JSON (no reasoning, no extra text):\n\n"
@@ -91,7 +98,7 @@ def execute(self, state: WorkflowState) -> Dict[str, Any]:
             strength=strength,
             rationale=rec_dict["rationale"],
             caveats=rec_dict.get("caveats", []),
-            evidence_quality=evidence_quality
+            evidence_quality=evidence_quality,
         )
 
         return {"recommendation": recommendation}

src/agents/appraise_agent.py

@@ -12,16 +12,16 @@
 # Initial GRADE points by study type (4=High, 3=Moderate, 2=Low, 1=Very Low)
 _INITIAL_POINTS: Dict[str, int] = {
     "RCT": 4,
-    "SYSTEMATIC_REVIEW": 4,   # Starts High (synthesizes RCTs or best available evidence)
-    "META_ANALYSIS": 4,        # Starts High
-    "NMA": 4,                  # Network meta-analysis: starts High
+    "SYSTEMATIC_REVIEW": 4,  # Starts High (synthesizes RCTs or best available evidence)
+    "META_ANALYSIS": 4,  # Starts High
+    "NMA": 4,  # Network meta-analysis: starts High
     "COHORT": 2,
     "CASE_CONTROL": 2,
-    "CROSS_SECTIONAL": 2,      # Observational: starts Low
-    "NARRATIVE_REVIEW": 1,     # Expert synthesis without systematic search: Very Low
+    "CROSS_SECTIONAL": 2,  # Observational: starts Low
+    "NARRATIVE_REVIEW": 1,  # Expert synthesis without systematic search: Very Low
     "CASE_REPORT": 1,
-    "GUIDELINE": 3,            # Typically based on SR: starts Moderate
-    "EXPERT_OPINION": 1,       # No systematic search: Very Low
+    "GUIDELINE": 3,  # Typically based on SR: starts Moderate
+    "EXPERT_OPINION": 1,  # No systematic search: Very Low
 }
 
 # Mapping from GRADE codes to human-readable study type labels
@@ -123,7 +123,9 @@ def __init__(self, llm, tools: List[Any] = None):
         self.prompt_template = self._load_prompt()
 
     def _load_prompt(self) -> str:
-        prompt_path = Path(__file__).parent.parent / "config" / "prompts" / "appraise_agent.txt"
+        prompt_path = (
+            Path(__file__).parent.parent / "config" / "prompts" / "appraise_agent.txt"
+        )
         with open(prompt_path, "r", encoding="utf-8") as f:
             return f.read()
 
@@ -137,7 +139,11 @@ def _format_evidence_list(self, evidence_list) -> str:
         for i, e in enumerate(evidence_list):
             abstract_preview = (getattr(e, "abstract", "") or "")[:200]
             study_type_hint = getattr(e, "study_type", "") or ""
-            hint_str = f"\nSource DB study_type hint: {study_type_hint}" if study_type_hint else ""
+            hint_str = (
+                f"\nSource DB study_type hint: {study_type_hint}"
+                if study_type_hint
+                else ""
+            )
             parts.append(
                 f"Evidence {i + 1}:\n"
                 f"Title: {e.title}\n"
@@ -157,7 +163,9 @@ def _appraise_batch(self, evidence_subset: list, backtrack_context: str) -> dict
         response = self.llm.invoke(prompt)
         return self._parse_json(response.content)
 
-    def _merge_appraisal_dicts(self, results: List[dict], batch_sizes: List[int]) -> dict:
+    def _merge_appraisal_dicts(
+        self, results: List[dict], batch_sizes: List[int]
+    ) -> dict:
         """
         Merge multiple batch appraisal results into a single appraisal dict.
 
@@ -175,8 +183,12 @@ def _merge_appraisal_dicts(self, results: List[dict], batch_sizes: List[int]) ->
                 global_id += 1
 
         # Conflict: pessimistic — YES wins
-        merged_conflict = {"conflicts_exist": "NO", "conflict_type": "NA",
-                           "conflict_severity": "NA", "conflict_description": None}
+        merged_conflict = {
+            "conflicts_exist": "NO",
+            "conflict_type": "NA",
+            "conflict_severity": "NA",
+            "conflict_description": None,
+        }
         for result in results:
             c = result.get("conflict_assessment", {})
             if c.get("conflicts_exist") == "YES":
@@ -194,7 +206,9 @@ def _merge_appraisal_dicts(self, results: List[dict], batch_sizes: List[int]) ->
                 best_numerical = n
 
         # Summary: join all
-        summaries = [r.get("evidence_summary", "") for r in results if r.get("evidence_summary")]
+        summaries = [
+            r.get("evidence_summary", "") for r in results if r.get("evidence_summary")
+        ]
         merged_summary = " | ".join(summaries)
 
         return {
@@ -235,8 +249,12 @@ def execute(self, state: WorkflowState) -> Dict[str, Any]:
                     for batch in batches
                 ]
                 batch_results = [f.result() for f in futures]
-            print(f"[TIMING] Appraise parallel batches ({len(batches)}×{_BATCH_SIZE}): {time.time()-t0:.1f}s")
-            appraisal_dict = self._merge_appraisal_dicts(batch_results, [len(b) for b in batches])
+            print(
+                f"[TIMING] Appraise parallel batches ({len(batches)}×{_BATCH_SIZE}): {time.time()-t0:.1f}s"
+            )
+            appraisal_dict = self._merge_appraisal_dicts(
+                batch_results, [len(b) for b in batches]
+            )
             print(f"[PARALLEL-APPRAISE] {len(batches)} batches completed in parallel.")
         else:
             # Single batch (≤5 articles): no need for parallel overhead
@@ -260,27 +278,33 @@ def execute(self, state: WorkflowState) -> Dict[str, Any]:
             graded_evidence.append(evidence)
 
             # Build rich rationale record for downstream consumers (including Judge)
-            initial_grade = _POINTS_TO_GRADE.get(_INITIAL_POINTS.get(study_type, 1), "Very Low")
-            grade_rationales.append({
-                "evidence_id": i + 1,
-                "title": evidence.title,
-                "study_type": study_type,
-                "initial_grade": initial_grade,
-                "risk_of_bias": appraisal.get("risk_of_bias", "NOT_SERIOUS"),
-                "inconsistency": appraisal.get("inconsistency", "NA"),
-                "indirectness": appraisal.get("indirectness", "NOT_SERIOUS"),
-                "imprecision": appraisal.get("imprecision", "NOT_SERIOUS"),
-                "publication_bias": appraisal.get("publication_bias", "UNDETECTED"),
-                "large_effect": appraisal.get("large_effect", "NA"),
-                "dose_response": appraisal.get("dose_response", "NA"),
-                "computed_grade": computed_grade,
-                "rationale": appraisal.get("rationale", ""),
-            })
+            initial_grade = _POINTS_TO_GRADE.get(
+                _INITIAL_POINTS.get(study_type, 1), "Very Low"
+            )
+            grade_rationales.append(
+                {
+                    "evidence_id": i + 1,
+                    "title": evidence.title,
+                    "study_type": study_type,
+                    "initial_grade": initial_grade,
+                    "risk_of_bias": appraisal.get("risk_of_bias", "NOT_SERIOUS"),
+                    "inconsistency": appraisal.get("inconsistency", "NA"),
+                    "indirectness": appraisal.get("indirectness", "NOT_SERIOUS"),
+                    "imprecision": appraisal.get("imprecision", "NOT_SERIOUS"),
+                    "publication_bias": appraisal.get("publication_bias", "UNDETECTED"),
+                    "large_effect": appraisal.get("large_effect", "NA"),
+                    "dose_response": appraisal.get("dose_response", "NA"),
+                    "computed_grade": computed_grade,
+                    "rationale": appraisal.get("rationale", ""),
+                }
+            )
 
         # --- Conflict assessment ---
         conflict = appraisal_dict.get("conflict_assessment", {})
         has_conflict = conflict.get("conflicts_exist") == "YES"
-        conflict_description = conflict.get("conflict_description") if has_conflict else None
+        conflict_description = (
+            conflict.get("conflict_description") if has_conflict else None
+        )
 
         # --- Numerical assessment ---
         numerical = appraisal_dict.get("numerical_assessment", {})
@@ -304,5 +328,6 @@ def execute(self, state: WorkflowState) -> Dict[str, Any]:
                 "note": numerical.get("note", ""),
             },
             # True only when there is a MAJOR conflict (influences scheduling)
-            "bias_inconsistency": has_conflict and conflict.get("conflict_severity") == "MAJOR",
+            "bias_inconsistency": has_conflict
+            and conflict.get("conflict_severity") == "MAJOR",
         }

src/agents/ask_agent.py

@@ -3,6 +3,7 @@
 from src.agents.base import BaseAgent, robust_parse_json
 from src.state.schema import WorkflowState, PICOQuery
 
+
 class AskAgent(BaseAgent):
     """Agent for refining clinical questions into PICO format"""
 
@@ -12,8 +13,10 @@ def __init__(self, llm, tools: List[Any] = None):
 
     def _load_prompt(self) -> str:
         """Load prompt template from file"""
-        prompt_path = Path(__file__).parent.parent / "config" / "prompts" / "ask_agent.txt"
-        with open(prompt_path, 'r', encoding='utf-8') as f:
+        prompt_path = (
+            Path(__file__).parent.parent / "config" / "prompts" / "ask_agent.txt"
+        )
+        with open(prompt_path, "r", encoding="utf-8") as f:
             return f.read()
 
     def _parse_json(self, content: str) -> dict:
@@ -29,8 +32,7 @@ def execute(self, state: WorkflowState) -> Dict[str, Any]:
             backtrack_context = f"Previous attempt failed: {state['backtrack_reason']}\nPlease refine the question."
 
         prompt = self.prompt_template.format(
-            question=question,
-            backtrack_context=backtrack_context
+            question=question, backtrack_context=backtrack_context
         )
 
         response = self.llm.invoke(prompt)
@@ -41,7 +43,7 @@ def execute(self, state: WorkflowState) -> Dict[str, Any]:
             intervention=pico_dict["intervention"],
             comparison=pico_dict["comparison"],
             outcome=pico_dict["outcome"],
-            keywords=pico_dict["keywords"]
+            keywords=pico_dict["keywords"],
         )
 
         question_type = pico_dict.get("question_type", "Therapy")