[Feature & Fix] Enhance session and history search and analysis

README.md

@@ -5,12 +5,14 @@
 
 ### News
 
+- **2026-08-12**: v0.0.9 released — TUI thinking display (LLM reasoning surfaced in-place with spinner preview + final panel), approval bypass mode (`approval_bypass` config + session-wide "Allow ALL"), workspace boundary softened to approval-gated, long-task convergence hardening (false-progress fix, repeated-read gate, periodic checkpoint forcing, pre-compression knowledge extraction), cross-session task history (automatic session summaries + proactive history injection), dynamic tool registry rebuild for late-registered tools, terminal sessions enabled by default.
 - **2026-08-06**: v0.0.8 released — Cross-platform Windows support (DaemonTransport protocol with TCP loopback IPC), real journey test layer with cassette-backed CI (6 e2e journeys, cost-bounded), community Windows fixes (@fanqiNO1). 1,540 tests.
-- **2026-08-06**: v0.0.7 released — 1M-class context windows end-to-end, self-calibrating token estimator, internal-defect failure category, concurrent-TUI session identity isolation. 1,442 tests.
 
 <details>
 <summary>Previous releases</summary>
 
+- **2026-08-06**: v0.0.7 released — 1M-class context windows end-to-end, self-calibrating token estimator, internal-defect failure category, concurrent-TUI session identity isolation. 1,442 tests.
+
 - **2026-07-31**: v0.0.6 released — side-effect-gated recovery (checkpointed halts with structured `InteractionRequest`), uncertain-effect reporting for failed outbound calls, centralized logging with an independent daemon log level, session-bound LeapBoard analysis, platform-neutral gateway validators, and end-to-end architecture contract tests with the CI gate restored.
 - **2026-07-28**: v0.0.5 released — adaptive-depth execution for long-horizon tasks, built-in coding tools, per-session daemon concurrency, improved TUI stability, and hardened leapd recovery/status diagnostics.
 - **2026-07-16**: **LeapBoard** — general-purpose monitoring web dashboard (Watch→Finding + Server-Driven UI); `/board` entry, live session analysis, and finance/sentiment/research templates.

src/leapflow/cli/commands/interactive.py

@@ -539,6 +539,10 @@ async def _stream_response(prompt_text: str) -> None:
                         renderer.feed(event.content)
                     elif event.type == "thinking":
                         renderer.feed_thinking(event.content)
+                        # Phase 2: route thinking to spinner for in-place display
+                        _preview = (event.content or "").strip().replace("\n", " ")[:80]
+                        if _preview:
+                            app.spinner_text = f"💭 {_preview}"
                     elif event.type == "tool_start":
                         app.spinner_text = renderer.tool_started(
                             event.content,
@@ -1213,6 +1217,10 @@ async def _stream_response(
                     renderer.feed(event.content)
                 elif event.type == "thinking":
                     renderer.feed_thinking(event.content)
+                    # Phase 2: route thinking to spinner for in-place display
+                    _preview = (event.content or "").strip().replace("\n", " ")[:80]
+                    if _preview:
+                        app.spinner_text = f"💭 {_preview}"
                 elif event.type == "tool_start":
                     app.spinner_text = renderer.tool_started(
                         event.content,

src/leapflow/cli/context.py

@@ -1806,6 +1806,56 @@ async def _session_search_handler(params: dict) -> dict:
                 TOOL_HANDLERS["session_search"] = _session_search_handler
                 TOOL_HANDLERS["gp_session_search"] = _session_search_handler
                 logger.debug("session_search tool registered")
+
+                # ── Register session_list tool ──
+                def _format_ts(ts: float) -> str:
+                    if not ts:
+                        return ""
+                    import datetime
+                    try:
+                        return datetime.datetime.fromtimestamp(ts).strftime("%Y-%m-%d %H:%M")
+                    except (TypeError, ValueError, OSError):
+                        return str(ts)[:16]
+
+                TOOL_DEFINITIONS.append({
+                    "type": "function",
+                    "function": {
+                        "name": "session_list",
+                        "description": (
+                            "List recent conversation sessions with titles, dates, and summaries. "
+                            "Use for browsing past tasks or when user asks to see history without specific search terms. "
+                            "No keywords needed — returns chronological list."
+                        ),
+                        "parameters": {
+                            "type": "object",
+                            "properties": {
+                                "limit": {"type": "integer", "description": "Max sessions to return (default: 10, max: 30)"},
+                            },
+                            "required": [],
+                        },
+                    },
+                })
+
+                async def _session_list_handler(params: dict) -> dict:
+                    limit = min(int(params.get("limit", 10)), 30)
+                    sessions = conv_store.list_sessions(limit=limit, active_only=False)
+                    items = []
+                    for s in sessions:
+                        item = {
+                            "title": getattr(s, 'title', '') or getattr(s, 'session_id', '')[:8],
+                            "date": _format_ts(getattr(s, 'updated_at', 0) or getattr(s, 'created_at', 0)),
+                            "messages": getattr(s, 'message_count', 0),
+                        }
+                        summary = getattr(s, 'summary', '')
+                        if summary:
+                            item["summary"] = summary[:200]
+                        items.append(item)
+                    import json as _json_sl
+                    return {"ok": True, "result": _json_sl.dumps(items, ensure_ascii=False)}
+
+                TOOL_HANDLERS["session_list"] = _session_list_handler
+                TOOL_HANDLERS["gp_session_list"] = _session_list_handler
+                logger.debug("session_list tool registered")
             except Exception:
                 logger.debug("session_search tool registration failed", exc_info=True)
 
@@ -2588,7 +2638,7 @@ async def _persist_session_summary(self) -> None:
             conv_store = getattr(self, '_conversation_store', None)
             if conv_store and session_id and goal_line:
                 try:
-                    conv_store.end_session(session_id, title=goal_line)
+                    conv_store.end_session(session_id, title=goal_line, summary=summary[:500])
                 except Exception:
                     logger.debug("session title update failed", exc_info=True)
         except Exception:

src/leapflow/cli/tui_app/stream.py

@@ -430,8 +430,10 @@ def tool_started(self, name: str, metadata: dict[str, Any] | None = None) -> str
         Discards any pending content — it was preamble preceding the tool call
         and should not appear in the final answer.
         """
-        # Flush accumulated thinking before starting new tool display
-        if self._thinking_buffer.strip():
+        # Flush thinking ONCE per LLM round: only when this is the first tool
+        # in a new batch (no other tools currently active). This shows one
+        # thinking panel per round during execution, not per individual tool.
+        if not self._active_tools and self._thinking_buffer.strip():
             self._console.thinking(self._thinking_buffer)
             self._thinking_buffer = ""
         self._pending = ""

src/leapflow/engine/engine.py

@@ -980,8 +980,17 @@ def _extract_json_object(text: str) -> Dict[str, Any]:
 
 
 def _keywords_from_query(q: str) -> list[str]:
-    toks = re.findall(r"[\w\-./]+|[\u4e00-\u9fff]+", q)
-    return [t for t in toks if len(t) >= 2][:12]
+    tokens: list[str] = []
+    for segment in re.findall(r'[\u4e00-\u9fff]+|[\w\-./]+', q):
+        if re.match(r'[\u4e00-\u9fff]', segment):
+            if len(segment) == 1:
+                tokens.append(segment)
+            else:
+                for i in range(len(segment) - 1):
+                    tokens.append(segment[i:i+2])
+        elif len(segment) >= 2:
+            tokens.append(segment)
+    return tokens[:12]
 
 
 @dataclass(frozen=True, slots=True)

src/leapflow/memory/manager.py

@@ -9,6 +9,7 @@
 import json
 import logging
 import math
+import re as _re
 from pathlib import Path
 from typing import Any, Dict, List, Optional
 
@@ -24,6 +25,26 @@
 logger = logging.getLogger(__name__)
 
 
+# ──────────────────────────────────────────────────────────────────────
+# CJK-aware tokenization for search
+# ──────────────────────────────────────────────────────────────────────
+
+def _tokenize_for_search(text: str) -> list[str]:
+    """Split text into search tokens. CJK runs become overlapping bigrams."""
+    tokens: list[str] = []
+    for segment in _re.findall(r'[\u4e00-\u9fff]+|[a-zA-Z0-9_\-./]+', text):
+        if _re.match(r'[\u4e00-\u9fff]', segment):
+            # CJK: overlapping bigrams (or single char if length 1)
+            if len(segment) == 1:
+                tokens.append(segment)
+            else:
+                for i in range(len(segment) - 1):
+                    tokens.append(segment[i:i+2])
+        else:
+            tokens.append(segment.lower())
+    return tokens[:8]
+
+
 # ──────────────────────────────────────────────────────────────────────
 # Shared decay formula (avoids circular import)
 # ──────────────────────────────────────────────────────────────────────
@@ -296,7 +317,7 @@ async def prefetch(
         session_scope: str = "",
     ) -> List[MemoryEntry]:
         """Quick search for LLM context injection with optional project/task/session scope."""
-        keywords = query_text.split()[:5]
+        keywords = _tokenize_for_search(query_text)
         scope_terms = [term for term in (scope_keywords or []) if term]
         query = MemoryQuery(
             keywords=[*keywords, *scope_terms[:5]],
@@ -628,7 +649,7 @@ async def _handle_memory_search(
 
         domains = [SignalDomain(domain_str)] if domain_str else None
         mq = MemoryQuery(
-            keywords=query_text.split()[:8],
+            keywords=_tokenize_for_search(query_text),
             domains=domains,
             limit=limit,
             workspace_root=workspace_root,

src/leapflow/memory/providers/semantic.py

@@ -176,12 +176,13 @@ async def search(self, query: MemoryQuery) -> List[MemoryEntry]:
             conditions.append("session_id = ?")
             params.append(query.session_scope)
 
-        # Keyword filter (AND semantics)
-        keywords = [k.strip() for k in query.keywords if k.strip()] if query.keywords else []
+        # Keyword filter (OR semantics — broadens recall for CJK bigrams)
+        keywords = [k.strip() for k in query.keywords if k.strip()][:8] if query.keywords else []
         if keywords:
-            for _ in keywords:
-                conditions.append("content ILIKE ?")
-                params.append(f"%{_}%")
+            or_clause = " OR ".join(["content ILIKE ?"] * len(keywords))
+            conditions.append(f"({or_clause})")
+            for kw in keywords:
+                params.append(f"%{kw}%")
 
         # Kind filter
         if query.kinds:

src/leapflow/prompts/templates.py

@@ -114,7 +114,10 @@ def build_react_system(language: str = "en", skill_catalog: str = "") -> str:
 5. **Handle failures gracefully**: If a tool fails, explain what went wrong and suggest alternatives. If it failed because of an unknown tool/platform/action name, retry once with an exact name from the returned suggestions, then explain rather than keep guessing.
 6. **Summarize results naturally**: After tool execution, synthesize the results into a helpful answer rather than dumping raw output.
 7. **Stay conversational**: Maintain a natural, helpful tone. Acknowledge context from earlier in the conversation.
-8. **Recall past work**: When the user references prior tasks or asks about past work, use session_search to recall relevant context from previous sessions.
+8. **Recall past work**:
+   - Broad queries ("之前做了什么", "列出任务"): answer from the "Recent Task History" section already in your context. If insufficient, call session_list.
+   - Specific lookups ("上次那个配置怎么改的"): call session_search with relevant phrases (NOT single characters).
+   - Do NOT call search tools repeatedly with keyword fragments. One well-phrased call is better than ten fragmented ones.
 
 ## Coding & Verification
 When working with code, prefer the precise built-in tools over ad-hoc shell: use `repo_map` to orient in an