Merge pull request #17 from boostcampwm-snu-2026-1/dev

rlafurud · web-flow · commit 4b317eec8a73 · 2026-06-05T17:17:25.000+09:00
Release: dev → main (CI, dead code 정리, fallback, 에러/로딩 UX)
diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
@@ -0,0 +1,44 @@
+name: CI
+
+on:
+  push:
+    branches: [main, dev]
+  pull_request:
+    branches: [main, dev]
+
+jobs:
+  frontend:
+    name: Frontend build
+    runs-on: ubuntu-latest
+    defaults:
+      run:
+        working-directory: frontend
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-node@v4
+        with:
+          node-version: '20'
+          cache: npm
+          cache-dependency-path: frontend/package-lock.json
+      - run: npm ci
+      - run: npm run build
+
+  backend:
+    name: Backend import & compile
+    runs-on: ubuntu-latest
+    defaults:
+      run:
+        working-directory: backend
+    env:
+      # dummy key so module import (which builds no client) is safe in CI
+      OPENAI_API_KEY: sk-ci-dummy
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-python@v5
+        with:
+          python-version: '3.11'
+      - run: pip install -r requirements.txt
+      - name: Compile all modules
+        run: python -m compileall -q .
+      - name: Import API app
+        run: python -c "import app; print('app import OK')"
diff --git a/backend/pipeline.py b/backend/pipeline.py
@@ -30,15 +30,57 @@
 # --------------------------------------------------------------------------- #
 # (1) Find the most similar existing scenario in the DB
 # --------------------------------------------------------------------------- #
+def _scenario_details() -> dict:
+    try:
+        with open(DB_DIR / "scenario" / "meta.json", "r", encoding="utf-8") as f:
+            return json.load(f).get("details", {})
+    except Exception:
+        return {}
+
+
+def _keyword_fallback(query: str, details: dict) -> str:
+    """Pick the scenario whose name+description shares the most words with the
+    query. No LLM — used when the DBCall match is empty or errors out."""
+    q = set(re.findall(r"[a-z0-9가-힣]+", query.lower()))
+    best, best_score = None, -1
+    for name, info in details.items():
+        text = f"{name} {info.get('description', '')}".lower()
+        score = len(q & set(re.findall(r"[a-z0-9가-힣]+", text)))
+        if score > best_score:
+            best, best_score = name, score
+    return best if best_score > 0 else None
+
+
 def find_scenario(query: str, seed: int = None) -> str:
-    """Return the best-matching scenario name from db/scenario, or None."""
-    db = DBCall(seed=seed)
-    _, names = db.call_with_names(query, folder="scenario")
-    if not names:
-        print("   ⚠️ No matching scenario found in DB.")
-        return None
-    print(f"   🔎 Matched scenario(s): {names} -> using '{names[0]}'")
-    return names[0]
+    """Return the best-matching scenario name from db/scenario.
+
+    Resilient: LLM match → keyword fallback → default to the first scenario.
+    Returns None only when the scenario DB is empty/unreadable.
+    """
+    details = _scenario_details()
+    names = []
+    try:
+        db = DBCall(seed=seed)
+        _, names = db.call_with_names(query, folder="scenario")
+    except Exception as e:
+        print(f"   ⚠️ DBCall match failed ({e}); falling back.")
+
+    if names:
+        print(f"   🔎 Matched scenario(s): {names} -> using '{names[0]}'")
+        return names[0]
+
+    fb = _keyword_fallback(query, details)
+    if fb:
+        print(f"   ↩️ Keyword fallback -> '{fb}'")
+        return fb
+
+    if details:
+        first = next(iter(details))
+        print(f"   ↩️ No overlap; defaulting to first scenario -> '{first}'")
+        return first
+
+    print("   ⚠️ Scenario DB is empty.")
+    return None
 
 
 # --------------------------------------------------------------------------- #
diff --git a/backend/script_builder.py b/backend/script_builder.py
@@ -9,7 +9,7 @@
 from langchain_core.output_parsers import JsonOutputParser, StrOutputParser
 from dotenv import load_dotenv
 from pathlib import Path
-from developer_prompt import SELECT_MAP_PROMPT, PLACE_UNITS_PROMPT, RULE_CONFIG_PROMPT, GET_CONDITION_PROMPT, REFINE_SCRIPT_PROMPT
+from developer_prompt import SELECT_MAP_PROMPT, PLACE_UNITS_PROMPT, RULE_CONFIG_PROMPT, GET_CONDITION_PROMPT
 from db_call import DBCall
 
 import common
@@ -413,156 +413,6 @@ def assemble_draft(self, state: ScriptDeveloperState) -> Dict:
             "loop_count": 1  # Start at 1 so initial verify + MAX_LOOPS-1 refines = 3 total
         }
     
-    def verify_script(self, state: ScriptDeveloperState) -> Dict:
-        print("\n Verifying Script with Analyst...")
-        current_json = state.get("final_json", {})
-
-        analyst_input_state = {
-            "current_script": current_json,
-            "current_phase": "compiler",
-            "loop_count": state.get("loop_count", 0),
-            "target_diffs": state.get("refined_diffs", None)
-        }
-        
-        try:
-            # Call Analyst's verification logic
-            result = self.analyst.verify_script(analyst_input_state)
-            
-            is_valid = result.get("game_script_valid", False)
-            feedback_list = result.get("game_script_feedback", [])
-            
-            # Format feedback for LLM if it's a list
-            feedback_str = "\n".join(feedback_list) if isinstance(feedback_list, list) else str(feedback_list)
-
-            if is_valid:
-                print("   ✅ Analyst: Script is VALID.")
-                # Log successful validation
-                self._log_feedback(
-                    feedback="Script validation passed",
-                    is_valid=True,
-                    loop_count=state.get("loop_count", 0),
-                    step_num=getattr(self, 'current_step_num', None)
-                )
-            else:
-                print(f"   ❌ Analyst: Script is INVALID.\n   Errors: {feedback_str}")
-                # Log failed validation
-                self._log_feedback(
-                    feedback=feedback_str,
-                    is_valid=False,
-                    loop_count=state.get("loop_count", 0),
-                    step_num=getattr(self, 'current_step_num', None)
-                )
-
-            return {
-                "is_script_valid": is_valid,
-                "script_feedback": feedback_str
-            }
-
-        except Exception as e:
-            print(f"   ⚠️ Analyst Verification Error: {e}")
-            return {"is_script_valid": False, "script_feedback": str(e)}
-        
-    def refine_script(self, state: ScriptDeveloperState) -> Dict:
-        print("\n[ScriptDev] 🔧 Refining Script...")
-
-        current_json = state.get("final_json", {})
-        feedback = state.get("script_feedback", "")
-        user_intent = state.get("user_intent", "")
-        loop_count = state.get("loop_count", 0) + 1
-
-        # Single difficulty — always refine "normal"
-        failed_diffs = {"normal"}
-
-        prompt = ChatPromptTemplate.from_messages([
-            ("system", REFINE_SCRIPT_PROMPT),
-            ("user", """
-            [User Intent]: {user_intent}
-            [Error Feedback]: {feedback}
-            [Current JSON]: {current_json}
-            """)
-        ])
-        chain = prompt | self.client | self.parser
-
-        merged_json = dict(current_json)
-        any_fixed = False
-
-        for diff in failed_diffs:
-            if diff not in current_json:
-                continue
-            diff_feedback = "\n".join(
-                line for line in feedback.splitlines()
-                if f"[{diff.upper()}]" in line.upper() or f"[{diff}]" in line.lower()
-            ) or feedback  # fallback to full feedback if no per-diff lines
-
-            print(f"   Fixing [{diff.upper()}]...")
-            try:
-                fixed = chain.invoke({
-                    "user_intent": user_intent,
-                    "feedback": diff_feedback,
-                    "current_json": json.dumps(current_json[diff], ensure_ascii=False)
-                })
-                if isinstance(fixed, dict):
-                    merged_json[diff] = fixed
-                    any_fixed = True
-                    print(f"   ✅ [{diff.upper()}] Fixed")
-                else:
-                    print(f"   ⚠️ [{diff.upper()}] LLM returned non-dict, skipping")
-            except Exception as e:
-                print(f"   ❌ [{diff.upper()}] Fix failed: {e}")
-
-        if any_fixed:
-            print(f"   ✅ Script Refined (Loop {loop_count})")
-        else:
-            print(f"   ⚠️ No difficulties were fixed (Loop {loop_count})")
-
-        return {
-            "final_json": merged_json,
-            "loop_count": loop_count,
-            "is_script_valid": False,
-            "refined_diffs": list(failed_diffs)
-        }
-    
-    def _log_feedback(self, feedback: str, is_valid: bool, loop_count: int, step_num: int = None):
-        """Log feedback for script validation/refinement."""
-        feedback_entry = {
-            "timestamp": datetime.datetime.now().isoformat(),
-            "source": "script_developer",
-            "phase": "script_developer",
-            "loop_count": loop_count,
-            "is_valid": is_valid,
-            "feedback": feedback
-        }
-        if step_num is not None:
-            feedback_entry["step_num"] = step_num
-
-        self.feedback_logs.append(feedback_entry)
-
-        # Save immediately
-        try:
-            with open(self.feedback_log_path, "w", encoding="utf-8") as f:
-                json.dump(self.feedback_logs, f, indent=2, ensure_ascii=False)
-        except Exception as e:
-            print(f"       ⚠️ Failed to save feedback log: {e}")
-
-    def check_script_validity(self, state: ScriptDeveloperState):
-        is_valid = state.get("is_script_valid", False)
-        loop_count = state.get("loop_count", 0)
-        MAX_LOOPS = 3
-
-        if is_valid:
-            return "valid"
-        elif loop_count >= MAX_LOOPS:
-            print(f"   ⚠️ Max refine loops ({MAX_LOOPS}) reached. Returning to router.")
-            self._log_feedback(
-                feedback=f"Max refine loops ({MAX_LOOPS}) reached — returning to router",
-                is_valid=False,
-                loop_count=loop_count,
-                step_num=getattr(self, 'current_step_num', None)
-            )
-            return "max_retries"
-        else:
-            return "invalid"
-
     def run(self, user_intent: str, gdd: Dict, meta_feedback: str = None, step_num: int = None) -> Dict[str, Any]:
         self.current_step_num = step_num  # Store for use in other methods
         print(f"\n[ScriptDev] 🚀 Starting Script Generation Workflow for: '{user_intent}' (Step {step_num})")
diff --git a/frontend/src/App.jsx b/frontend/src/App.jsx
@@ -6,11 +6,25 @@ import MiniMap from './components/MiniMap'
 import SimPlayback from './components/SimPlayback'
 import JsonView from './components/JsonView'
 
+// Map backend error codes / raw messages to friendly Korean text.
+function friendlyError(err) {
+  const map = {
+    no_matching_scenario: '비슷한 시나리오를 찾지 못했어요. 프롬프트를 조금 더 구체적으로 적어보세요.',
+    failed_to_load_scenario: '시나리오 데이터를 불러오지 못했어요. 잠시 후 다시 시도해주세요.',
+  }
+  if (map[err]) return map[err]
+  if (typeof err === 'string' && /failed to fetch|networkerror|load failed/i.test(err)) {
+    return '백엔드에 연결하지 못했어요. 서버가 실행 중인지 확인해주세요 (localhost:8000).'
+  }
+  return `생성에 실패했어요: ${err}`
+}
+
 export default function App() {
   const [catalog, setCatalog] = useState({ scenarios: [], maps: [] })
   const [loading, setLoading] = useState(false)
   const [error, setError] = useState(null)
   const [result, setResult] = useState(null)
+  const [lastQuery, setLastQuery] = useState('')
 
   useEffect(() => {
     getCatalog()
@@ -19,18 +33,19 @@ export default function App() {
   }, [])
 
   async function handleGenerate(query) {
+    setLastQuery(query)
     setLoading(true)
     setError(null)
     setResult(null)
     try {
       const res = await generate(query)
       if (res.error || !res.config) {
-        setError(res.error || 'No config was generated for this prompt.')
+        setError(friendlyError(res.error || 'no_matching_scenario'))
       } else {
         setResult(res)
       }
     } catch (e) {
-      setError(e.message)
+      setError(friendlyError(e.message))
     } finally {
       setLoading(false)
     }
@@ -56,9 +71,23 @@ export default function App() {
         onGenerate={handleGenerate}
       />
 
-      {error && <div className="error">⚠️ {error}</div>}
+      {error && !loading && (
+        <div className="error">
+          <span>⚠️ {error}</span>
+          {lastQuery && (
+            <button className="btn retry" onClick={() => handleGenerate(lastQuery)}>
+              다시 시도
+            </button>
+          )}
+        </div>
+      )}
 
-      {loading && <div className="loading">생성 중… (DB 매칭 → 스크립트 작성)</div>}
+      {loading && (
+        <div className="loading">
+          <span className="spinner" aria-hidden="true" />
+          생성 중… (DB 매칭 → 스크립트 작성)
+        </div>
+      )}
 
       {result && config && (
         <div className="result">
diff --git a/frontend/src/styles.css b/frontend/src/styles.css
@@ -88,11 +88,23 @@ body {
   margin-top: 18px; padding: 14px 16px;
   background: #3b1414; border: 1px solid #7f1d1d; color: #fecaca;
   border-radius: 10px;
+  display: flex; align-items: center; justify-content: space-between; gap: 12px;
 }
+.error .retry {
+  background: #7f1d1d; border-color: #b91c1c; color: #fee2e2; white-space: nowrap;
+}
+.error .retry:hover { background: #991b1b; }
 .loading {
   margin-top: 18px; padding: 14px 16px; color: var(--muted);
   background: var(--panel); border: 1px solid var(--border); border-radius: 10px;
+  display: flex; align-items: center; gap: 10px;
+}
+.spinner {
+  width: 16px; height: 16px; border-radius: 50%;
+  border: 2px solid var(--border); border-top-color: var(--accent);
+  display: inline-block; animation: spin 0.7s linear infinite;
 }
+@keyframes spin { to { transform: rotate(360deg); } }
 
 /* result */
 .result { margin-top: 24px; display: flex; flex-direction: column; gap: 20px; }
