fix: improve merge logic, time decay, breath output format, session hook

.claude/hooks/session_breath.py

@@ -15,7 +15,6 @@
 #   OMBRE_HOOK_SKIP — set to "1" to disable the hook temporarily
 # ============================================================
 
-import json
 import os
 import sys
 import urllib.request
@@ -28,35 +27,18 @@ def main():
 
     base_url = os.environ.get("OMBRE_HOOK_URL", "http://localhost:8000").rstrip("/")
 
-    # Build MCP call via HTTP POST to the streamable-http endpoint
-    # The breath tool with no query triggers surfacing mode.
-    payload = json.dumps({
-        "jsonrpc": "2.0",
-        "id": 1,
-        "method": "tools/call",
-        "params": {
-            "name": "breath",
-            "arguments": {"query": "", "max_results": 2}
-        }
-    }).encode("utf-8")
-
     req = urllib.request.Request(
-        f"{base_url}/mcp",
-        data=payload,
-        headers={"Content-Type": "application/json"},
-        method="POST",
+        f"{base_url}/breath-hook",
+        headers={"Accept": "text/plain"},
+        method="GET",
     )
 
     try:
         with urllib.request.urlopen(req, timeout=8) as response:
             raw = response.read().decode("utf-8")
-            data = json.loads(raw)
-            # Extract text from MCP tool result
-            result_content = data.get("result", {}).get("content", [])
-            text_parts = [c.get("text", "") for c in result_content if c.get("type") == "text"]
-            output = "\n".join(text_parts).strip()
-            if output and output != "权重池平静，没有需要处理的记忆。":
-                print(f"[Ombre Brain - 记忆浮现]\n{output}")
+            output = raw.strip()
+            if output:
+                print(output)
     except (urllib.error.URLError, OSError):
         # Server not available (local stdio mode or not running) — silent fail
         pass

.claude/settings.json

@@ -7,7 +7,7 @@
           {
             "type": "command",
             "command": "python \"$CLAUDE_PROJECT_DIR/.claude/hooks/session_breath.py\"",
-            "shell": "powershell",
+            "shell": "/bin/bash",
             "timeout": 12,
             "statusMessage": "Ombre Brain 正在浮现记忆..."
           }
@@ -19,7 +19,7 @@
           {
             "type": "command",
             "command": "python \"$CLAUDE_PROJECT_DIR/.claude/hooks/session_breath.py\"",
-            "shell": "powershell",
+            "shell": "/bin/bash",
             "timeout": 12,
             "statusMessage": "Ombre Brain 正在浮现记忆..."
           }

README.md

@@ -166,8 +166,8 @@ Claude ←→ MCP Protocol ←→ server.py
 ### 步骤 / Steps
 
 ```bash
-git clone https://github.com/P0lar1zzZ/Ombre-Brain.git
-cd Ombre-Brain
+git clone https://git.p0lar1s.uk/P0lar1s/Ombre_Brain.git
+cd Ombre_Brain
 
 python -m venv .venv
 source .venv/bin/activate  # Windows: .venv\Scripts\activate

bucket_manager.py

@@ -87,8 +87,9 @@ class BucketManager:
         scoring = config.get("scoring_weights", {})
         self.w_topic = scoring.get("topic_relevance", 4.0)
         self.w_emotion = scoring.get("emotion_resonance", 2.0)
-        self.w_time = scoring.get("time_proximity", 1.5)
+        self.w_time = scoring.get("time_proximity", 2.5)
         self.w_importance = scoring.get("importance", 1.0)
+        self.content_weight = scoring.get("content_weight", 3.0)  # Added to allow better content-based matching during merge
 
     # ---------------------------------------------------------
     # Create a new bucket
@@ -573,9 +574,9 @@ class BucketManager:
             )
             * 2
         )
-        content_score = fuzz.partial_ratio(query, bucket.get("content", "")[:500]) * 1
+        content_score = fuzz.partial_ratio(query, bucket.get("content", "")[:1000]) * self.content_weight
 
-        return (name_score + domain_score + tag_score + content_score) / (100 * 8.5)
+        return (name_score + domain_score + tag_score + content_score) / (100 * 10.5)
 
     # ---------------------------------------------------------
     # Emotion resonance sub-score:
@@ -619,7 +620,7 @@ class BucketManager:
             days = max(0.0, (datetime.now() - last_active).total_seconds() / 86400)
         except (ValueError, TypeError):
             days = 30
-        return math.exp(-0.02 * days)
+        return math.exp(-0.1 * days)
 
     # ---------------------------------------------------------
     # List all buckets

check_buckets.py

@@ -0,0 +1,30 @@
+import asyncio
+from bucket_manager import BucketManager
+from utils import load_config
+
+async def main():
+    config = load_config()
+    bm = BucketManager(config)
+    buckets = await bm.list_all(include_archive=True)
+    
+    print(f"Total buckets: {len(buckets)}")
+    
+    domains = {}
+    for b in buckets:
+        for d in b.get("metadata", {}).get("domain", []):
+            domains[d] = domains.get(d, 0) + 1
+            
+    print(f"Domains: {domains}")
+    
+    # Check for formatting issues (e.g., missing critical fields)
+    issues = 0
+    for b in buckets:
+        meta = b.get("metadata", {})
+        if not meta.get("name") or not meta.get("domain") or not b.get("content"):
+            print(f"Format issue in {b['id']}")
+            issues += 1
+            
+    print(f"Found {issues} formatting issues.")
+
+if __name__ == "__main__":
+    asyncio.run(main())

dehydrator.py

@@ -193,21 +193,8 @@ class Dehydrator:
         if count_tokens_approx(content) < 100:
             return self._format_output(content, metadata)
 
-        # --- Try API compression first (best quality) ---
-        # --- 优先尝试 API 压缩 ---
-        if self.api_available:
-            try:
-                result = await self._api_dehydrate(content)
-                if result:
-                    return self._format_output(result, metadata)
-            except Exception as e:
-                logger.warning(
-                    f"API dehydration failed, degrading to local / "
-                    f"API 脱水失败，降级到本地压缩: {e}"
-                )
-
-        # --- Local compression fallback (works without API) ---
-        # --- 本地压缩兜底 ---
+        # --- Local compression (Always used as requested) ---
+        # --- 本地压缩 ---
         result = self._local_dehydrate(content)
         return self._format_output(result, metadata)
 
@@ -322,9 +309,10 @@ class Dehydrator:
         # --- Top-8 sentences + keyword list / 取高分句 + 关键词列表 ---
         selected = [s for _, s in scored[:8]]
         summary = "。".join(selected)
-        keyword_str = ", ".join(keywords[:10])
         
-        return f"[摘要] {summary}\n[关键词] {keyword_str}"
+        if len(summary) > 1000:
+            summary = summary[:1000] + "…"
+        return summary
 
     # ---------------------------------------------------------
     # Local merge (simple concatenation + truncation)
@@ -393,7 +381,6 @@ class Dehydrator:
         header = ""
         if metadata and isinstance(metadata, dict):
             name = metadata.get("name", "未命名")
-            tags = ", ".join(metadata.get("tags", []))
             domains = ", ".join(metadata.get("domain", []))
             try:
                 valence = float(metadata.get("valence", 0.5))
@@ -403,10 +390,10 @@ class Dehydrator:
             header = f"📌 记忆桶: {name}"
             if domains:
                 header += f" [主题:{domains}]"
-            if tags:
-                header += f" [标签:{tags}]"
             header += f" [情感:V{valence:.1f}/A{arousal:.1f}]"
             header += "\n"
+        
+        content = re.sub(r'\[\[([^\]]+)\]\]', r'\1', content)
         return f"{header}{content}"
 
     # ---------------------------------------------------------

server.py

@@ -89,6 +89,43 @@ async def health_check(request):
         return JSONResponse({"status": "error", "detail": str(e)}, status_code=500)
 
 
+# =============================================================
+# /breath-hook endpoint: Dedicated hook for SessionStart
+# 会话启动专用挂载点
+# =============================================================
+@mcp.custom_route("/breath-hook", methods=["GET"])
+async def breath_hook(request):
+    from starlette.responses import PlainTextResponse
+    try:
+        all_buckets = await bucket_mgr.list_all(include_archive=False)
+        # pinned
+        pinned = [b for b in all_buckets if b["metadata"].get("pinned") or b["metadata"].get("protected")]
+        # top 2 unresolved by score
+        unresolved = [b for b in all_buckets
+                      if not b["metadata"].get("resolved", False)
+                      and b["metadata"].get("type") != "permanent"
+                      and not b["metadata"].get("pinned")
+                      and not b["metadata"].get("protected")]
+        scored = sorted(unresolved, key=lambda b: decay_engine.calculate_score(b["metadata"]), reverse=True)
+        top = scored[:2]
+
+        parts = []
+        for b in pinned:
+            summary = await dehydrator.dehydrate(b["content"], {k: v for k, v in b["metadata"].items() if k != "tags"})
+            parts.append(f"📌 [核心准则] {summary}")
+        for b in top:
+            summary = await dehydrator.dehydrate(b["content"], {k: v for k, v in b["metadata"].items() if k != "tags"})
+            await bucket_mgr.touch(b["id"])
+            parts.append(summary)
+
+        if not parts:
+            return PlainTextResponse("")
+        return PlainTextResponse("[Ombre Brain - 记忆浮现]\n" + "\n---\n".join(parts))
+    except Exception as e:
+        logger.warning(f"Breath hook failed: {e}")
+        return PlainTextResponse("")
+
+
 # =============================================================
 # Internal helper: merge-or-create
 # 内部辅助：检查是否可合并，可以则合并，否则新建
@@ -111,7 +148,7 @@ async def _merge_or_create(
     返回 (桶ID或名称, 是否合并)。
     """
     try:
-        existing = await bucket_mgr.search(content, limit=1)
+        existing = await bucket_mgr.search(content, limit=1, domain_filter=domain or None)
     except Exception as e:
         logger.warning(f"Search for merge failed, creating new / 合并搜索失败，新建: {e}")
         existing = []
@@ -186,7 +223,8 @@ async def breath(
         pinned_results = []
         for b in pinned_buckets:
             try:
-                summary = await dehydrator.dehydrate(b["content"], b["metadata"])
+                clean_meta = {k: v for k, v in b["metadata"].items() if k != "tags"}
+                summary = await dehydrator.dehydrate(b["content"], clean_meta)
                 pinned_results.append(f"📌 [核心准则] {summary}")
             except Exception as e:
                 logger.warning(f"Failed to dehydrate pinned bucket / 钉选桶脱水失败: {e}")
@@ -211,7 +249,8 @@ async def breath(
         dynamic_results = []
         for b in top:
             try:
-                summary = await dehydrator.dehydrate(b["content"], b["metadata"])
+                clean_meta = {k: v for k, v in b["metadata"].items() if k != "tags"}
+                summary = await dehydrator.dehydrate(b["content"], clean_meta)
                 await bucket_mgr.touch(b["id"])
                 score = decay_engine.calculate_score(b["metadata"])
                 dynamic_results.append(f"[权重:{score:.2f}] {summary}")
@@ -249,7 +288,8 @@ async def breath(
     results = []
     for bucket in matches:
         try:
-            summary = await dehydrator.dehydrate(bucket["content"], bucket["metadata"])
+            clean_meta = {k: v for k, v in bucket["metadata"].items() if k != "tags"}
+            summary = await dehydrator.dehydrate(bucket["content"], clean_meta)
             await bucket_mgr.touch(bucket["id"])
             results.append(summary)
         except Exception as e:
@@ -271,7 +311,8 @@ async def breath(
                 drifted = random.sample(low_weight, min(random.randint(1, 3), len(low_weight)))
                 drift_results = []
                 for b in drifted:
-                    summary = await dehydrator.dehydrate(b["content"], b["metadata"])
+                    clean_meta = {k: v for k, v in b["metadata"].items() if k != "tags"}
+                    summary = await dehydrator.dehydrate(b["content"], clean_meta)
                     drift_results.append(f"[surface_type: random]\n{summary}")
                 results.append("--- 忽然想起来 ---\n" + "\n---\n".join(drift_results))
         except Exception as e: