refactor: add intent_classifier.py, register in init.py, LLM bridge shared

pipeline_service/init.py

@@ -367,5 +367,11 @@ def load_pipeline_service():
     # Load built-in interactive step types
     load_builtin_types()
 
+    # Register intent classifier + LLM bridge (shared across all pipelines)
+    from .intent_classifier import intent_classify
+    from .llm_bridge import llm_call
+    env.intent_classify = intent_classify
+    env.pipeline_llm_call = llm_call
+
     debug(f"[{MODULE_NAME}] v{MODULE_VERSION} loaded — pipeline engine with human-in-the-loop support")
     return True

pipeline_service/intent_classifier.py

@@ -0,0 +1,60 @@
+"""
+pipeline_service/intent_classifier.py — Reusable intent classification.
+
+DSPY can call: await intent_classify(model_name, message, intents, context)
+Registered on ServerEnv in pipeline_service/init.py
+"""
+
+import json
+import logging
+
+logger = logging.getLogger("pipeline.intent_classifier")
+
+INTENT_PROMPT = """你是一个意图分类器。分析用户输入，返回 JSON。
+
+可用意图：
+{intent_list}
+
+当前上下文：{context_str}
+
+返回纯 JSON（不要 markdown 包裹）：
+{{"intent":"...","confidence":0.0-1.0,"params":{{"key":"value"...}},"missing_info":"..."}}"""
+
+
+async def intent_classify(intents: list, message: str, context: dict = None) -> dict:
+    """Classify user intent. Returns {intent, confidence, params, missing_info}.
+
+    Args:
+        intents: [{"name":"new_project","description":"创建新项目"},...]
+        message: user input text
+        context: {"project_name":"...","iteration_name":"..."}
+    """
+    if not intents:
+        return {"intent": "chat", "confidence": 0.5, "params": {}, "missing_info": ""}
+
+    intent_list = "\n".join(
+        f"- {i['name']}: {i.get('description','')}" for i in intents
+    )
+    ctx_parts = []
+    if context:
+        for k, v in context.items():
+            if v:
+                ctx_parts.append(f"{k}={v}")
+    context_str = ", ".join(ctx_parts) if ctx_parts else "无上下文"
+
+    prompt = INTENT_PROMPT.format(intent_list=intent_list, context_str=context_str)
+
+    try:
+        from pipeline_service.llm_bridge import llm_call
+        raw = await llm_call(
+            prompt=f"{prompt}\n\n用户输入：{message}",
+            model=None,
+            temperature=0.2,
+        )
+        raw = raw.strip()
+        if raw.startswith("```"):
+            raw = raw.split("\n", 1)[1].rsplit("```", 1)[0]
+        return json.loads(raw)
+    except Exception as e:
+        logger.warning("intent_classify failed: %s", e)
+        return {"intent": "chat", "confidence": 0.3, "params": {}, "missing_info": ""}

pipeline_service/llm_bridge.py

@@ -17,7 +17,7 @@ _model_cache: dict = {}
 
 
 def _decrypt_key(encrypted: str) -> str:
-    """Decrypt api_key stored with password_encode."""
+    """Decrypt api_key stored with password_encode. Falls back to plaintext."""
     if not encrypted:
         return ""
     try:
@@ -27,7 +27,7 @@ def _decrypt_key(encrypted: str) -> str:
         key = config.password_key
         return unpassword(key, encrypted)
     except Exception:
-        return encrypted
+        return encrypted  # already plaintext or decrypt failed
 
 
 async def _get_model_config(model_name: str = None) -> dict: