feat(audit): 审计修复第四轮 — 跨会话搜索、LLM压缩集成、Presentation渲染器

- S9: MessageSearch 新增 Session/Global 双模式，Global 调用 VikingStorage memory_search - M4b: LLM 压缩器集成到 kernel AgentLoop，支持 use_llm 配置切换 - M4c: 压缩时自动提取记忆到 VikingStorage (runtime + tauri 双路径) - H6: 新增 ChartRenderer(recharts)、Document/Slideshow 完整渲染 - 累计修复 23 项，整体完成度 ~72%，真实可用率 ~80%
2026-03-27 11:44:14 +08:00
parent 7ae6990c97
commit 30b2515f07
16 changed files with 2121 additions and 245 deletions
--- a/crates/zclaw-runtime/src/compaction.rs
+++ b/crates/zclaw-runtime/src/compaction.rs
@@ -5,8 +5,18 @@
 //! token count exceeds the configured threshold, older messages are
 //! summarized into a single system message and only recent messages are
 //! retained.
+//!
+//! Supports two compaction modes:
+//! - **Rule-based**: Heuristic topic extraction (default, no LLM needed)
+//! - **LLM-based**: Uses an LLM driver to generate higher-quality summaries
+//!
+//! Optionally flushes old messages to the growth/memory system before discarding.

-use zclaw_types::Message;
+use std::sync::Arc;
+use zclaw_types::{AgentId, Message, SessionId};
+
+use crate::driver::{CompletionRequest, ContentBlock, LlmDriver};
+use crate::growth::GrowthIntegration;

 /// Number of recent messages to preserve after compaction.
 const DEFAULT_KEEP_RECENT: usize = 6;
@@ -146,6 +156,272 @@ pub fn maybe_compact(messages: Vec<Message>, threshold: usize) -> Vec<Message> {
    compacted
 }

+/// Configuration for compaction behavior.
+#[derive(Debug, Clone)]
+pub struct CompactionConfig {
+    /// Use LLM for generating summaries instead of rule-based extraction.
+    pub use_llm: bool,
+    /// Fall back to rule-based summary if LLM fails.
+    pub llm_fallback_to_rules: bool,
+    /// Flush memories from old messages before discarding them.
+    pub memory_flush_enabled: bool,
+    /// Maximum tokens for LLM-generated summary.
+    pub summary_max_tokens: u32,
+}
+
+impl Default for CompactionConfig {
+    fn default() -> Self {
+        Self {
+            use_llm: false,
+            llm_fallback_to_rules: true,
+            memory_flush_enabled: false,
+            summary_max_tokens: 500,
+        }
+    }
+}
+
+/// Outcome of an async compaction operation.
+#[derive(Debug, Clone)]
+pub struct CompactionOutcome {
+    /// The (possibly compacted) message list.
+    pub messages: Vec<Message>,
+    /// Number of messages removed during compaction.
+    pub removed_count: usize,
+    /// Number of memories flushed to the growth system.
+    pub flushed_memories: usize,
+    /// Whether LLM was used for summary generation.
+    pub used_llm: bool,
+}
+
+/// Async compaction with optional LLM summary and memory flushing.
+///
+/// When `messages` exceed `threshold` tokens:
+/// 1. If `memory_flush_enabled`, extract memories from old messages via growth system
+/// 2. Generate summary (LLM or rule-based depending on config)
+/// 3. Replace old messages with summary + keep recent messages
+pub async fn maybe_compact_with_config(
+    messages: Vec<Message>,
+    threshold: usize,
+    config: &CompactionConfig,
+    agent_id: &AgentId,
+    session_id: &SessionId,
+    driver: Option<&Arc<dyn LlmDriver>>,
+    growth: Option<&GrowthIntegration>,
+) -> CompactionOutcome {
+    let tokens = estimate_messages_tokens(&messages);
+    if tokens < threshold {
+        return CompactionOutcome {
+            messages,
+            removed_count: 0,
+            flushed_memories: 0,
+            used_llm: false,
+        };
+    }
+
+    tracing::info!(
+        "[Compaction] Triggered: {} tokens > {} threshold, {} messages",
+        tokens,
+        threshold,
+        messages.len(),
+    );
+
+    // Step 1: Flush memories from messages that are about to be compacted
+    let flushed_memories = if config.memory_flush_enabled {
+        if let Some(growth) = growth {
+            match growth
+                .process_conversation(agent_id, &messages, session_id.clone())
+                .await
+            {
+                Ok(count) => {
+                    tracing::info!(
+                        "[Compaction] Flushed {} memories before compaction",
+                        count
+                    );
+                    count
+                }
+                Err(e) => {
+                    tracing::warn!("[Compaction] Memory flush failed: {}", e);
+                    0
+                }
+            }
+        } else {
+            tracing::debug!("[Compaction] Memory flush requested but no growth integration available");
+            0
+        }
+    } else {
+        0
+    };
+
+    // Step 2: Determine split point (same logic as compact_messages)
+    let leading_system_count = messages
+        .iter()
+        .take_while(|m| matches!(m, Message::System { .. }))
+        .count();
+    let keep_from_end = DEFAULT_KEEP_RECENT
+        .min(messages.len().saturating_sub(leading_system_count));
+    let split_index = messages.len().saturating_sub(keep_from_end);
+    let split_index = split_index.max(leading_system_count);
+
+    if split_index == 0 {
+        return CompactionOutcome {
+            messages,
+            removed_count: 0,
+            flushed_memories,
+            used_llm: false,
+        };
+    }
+
+    let old_messages = &messages[..split_index];
+    let recent_messages = &messages[split_index..];
+    let removed_count = old_messages.len();
+
+    // Step 3: Generate summary (LLM or rule-based)
+    let summary = if config.use_llm {
+        if let Some(driver) = driver {
+            match generate_llm_summary(driver, old_messages, config.summary_max_tokens).await {
+                Ok(llm_summary) => {
+                    tracing::info!(
+                        "[Compaction] Generated LLM summary ({} chars)",
+                        llm_summary.len()
+                    );
+                    llm_summary
+                }
+                Err(e) => {
+                    if config.llm_fallback_to_rules {
+                        tracing::warn!(
+                            "[Compaction] LLM summary failed: {}, falling back to rules",
+                            e
+                        );
+                        generate_summary(old_messages)
+                    } else {
+                        tracing::warn!(
+                            "[Compaction] LLM summary failed: {}, returning original messages",
+                            e
+                        );
+                        return CompactionOutcome {
+                            messages,
+                            removed_count: 0,
+                            flushed_memories,
+                            used_llm: false,
+                        };
+                    }
+                }
+            }
+        } else {
+            tracing::warn!(
+                "[Compaction] LLM compaction requested but no driver available, using rules"
+            );
+            generate_summary(old_messages)
+        }
+    } else {
+        generate_summary(old_messages)
+    };
+
+    let used_llm = config.use_llm && driver.is_some();
+
+    // Step 4: Build compacted message list
+    let mut compacted = Vec::with_capacity(1 + recent_messages.len());
+    compacted.push(Message::system(summary));
+    compacted.extend(recent_messages.iter().cloned());
+
+    tracing::info!(
+        "[Compaction] Removed {} messages, {} remain (llm={})",
+        removed_count,
+        compacted.len(),
+        used_llm,
+    );
+
+    CompactionOutcome {
+        messages: compacted,
+        removed_count,
+        flushed_memories,
+        used_llm,
+    }
+}
+
+/// Generate a summary using an LLM driver.
+async fn generate_llm_summary(
+    driver: &Arc<dyn LlmDriver>,
+    messages: &[Message],
+    max_tokens: u32,
+) -> Result<String, String> {
+    let mut conversation_text = String::new();
+    for msg in messages {
+        match msg {
+            Message::User { content } => {
+                conversation_text.push_str(&format!("用户: {}\n", content))
+            }
+            Message::Assistant { content, .. } => {
+                conversation_text.push_str(&format!("助手: {}\n", content))
+            }
+            Message::System { content } => {
+                if !content.starts_with("[以下是之前对话的摘要]") {
+                    conversation_text.push_str(&format!("[系统]: {}\n", content))
+                }
+            }
+            Message::ToolUse { tool, input, .. } => {
+                conversation_text.push_str(&format!(
+                    "[工具调用 {}]: {}\n",
+                    tool.as_str(),
+                    input
+                ))
+            }
+            Message::ToolResult { output, .. } => {
+                conversation_text.push_str(&format!("[工具结果]: {}\n", output))
+            }
+        }
+    }
+
+    // Truncate conversation text if too long for the prompt itself
+    let max_conversation_chars = 8000;
+    if conversation_text.len() > max_conversation_chars {
+        conversation_text.truncate(max_conversation_chars);
+        conversation_text.push_str("\n...(对话已截断)");
+    }
+
+    let prompt = format!(
+        "请用简洁的中文总结以下对话的关键信息。保留重要的讨论主题、决策、结论和待办事项。\
+         输出格式为段落式摘要，不超过200字。\n\n{}",
+        conversation_text
+    );
+
+    let request = CompletionRequest {
+        model: String::new(),
+        system: Some(
+            "你是一个对话摘要助手。只输出摘要内容，不要添加额外解释。".to_string(),
+        ),
+        messages: vec![Message::user(&prompt)],
+        tools: Vec::new(),
+        max_tokens: Some(max_tokens),
+        temperature: Some(0.3),
+        stop: Vec::new(),
+        stream: false,
+    };
+
+    let response = driver
+        .complete(request)
+        .await
+        .map_err(|e| format!("{}", e))?;
+
+    // Extract text from content blocks
+    let text_parts: Vec<String> = response
+        .content
+        .iter()
+        .filter_map(|block| match block {
+            ContentBlock::Text { text } => Some(text.clone()),
+            _ => None,
+        })
+        .collect();
+
+    let summary = text_parts.join("");
+
+    if summary.is_empty() {
+        return Err("LLM returned empty response".to_string());
+    }
+
+    Ok(summary)
+}
+
 /// Generate a rule-based summary of old messages.
 fn generate_summary(messages: &[Message]) -> String {
    if messages.is_empty() {
--- a/crates/zclaw-runtime/src/lib.rs
+++ b/crates/zclaw-runtime/src/lib.rs
@@ -24,3 +24,4 @@ pub use loop_runner::{AgentLoop, AgentLoopResult, LoopEvent};
 pub use loop_guard::{LoopGuard, LoopGuardConfig, LoopGuardResult};
 pub use stream::{StreamEvent, StreamSender};
 pub use growth::GrowthIntegration;
+pub use compaction::{CompactionConfig, CompactionOutcome};
--- a/crates/zclaw-runtime/src/loop_runner.rs
+++ b/crates/zclaw-runtime/src/loop_runner.rs
@@ -12,7 +12,7 @@ use crate::tool::{ToolRegistry, ToolContext, SkillExecutor};
 use crate::tool::builtin::PathValidator;
 use crate::loop_guard::{LoopGuard, LoopGuardResult};
 use crate::growth::GrowthIntegration;
-use crate::compaction;
+use crate::compaction::{self, CompactionConfig};
 use zclaw_memory::MemoryStore;

 /// Agent loop runner
@@ -32,6 +32,8 @@ pub struct AgentLoop {
    growth: Option<GrowthIntegration>,
    /// Compaction threshold in tokens (0 = disabled)
    compaction_threshold: usize,
+    /// Compaction behavior configuration
+    compaction_config: CompactionConfig,
 }

 impl AgentLoop {
@@ -55,6 +57,7 @@ impl AgentLoop {
            path_validator: None,
            growth: None,
            compaction_threshold: 0,
+            compaction_config: CompactionConfig::default(),
        }
    }

@@ -115,6 +118,12 @@ impl AgentLoop {
        self
    }

+    /// Set compaction configuration (LLM mode, memory flushing, etc.)
+    pub fn with_compaction_config(mut self, config: CompactionConfig) -> Self {
+        self.compaction_config = config;
+        self
+    }
+
    /// Get growth integration reference
    pub fn growth(&self) -> Option<&GrowthIntegration> {
        self.growth.as_ref()
@@ -150,7 +159,23 @@ impl AgentLoop {

        // Apply compaction if threshold is configured
        if self.compaction_threshold > 0 {
-            messages = compaction::maybe_compact(messages, self.compaction_threshold);
+            let needs_async =
+                self.compaction_config.use_llm || self.compaction_config.memory_flush_enabled;
+            if needs_async {
+                let outcome = compaction::maybe_compact_with_config(
+                    messages,
+                    self.compaction_threshold,
+                    &self.compaction_config,
+                    &self.agent_id,
+                    &session_id,
+                    Some(&self.driver),
+                    self.growth.as_ref(),
+                )
+                .await;
+                messages = outcome.messages;
+            } else {
+                messages = compaction::maybe_compact(messages, self.compaction_threshold);
+            }
        }

        // Enhance system prompt with growth memories
@@ -316,7 +341,23 @@ impl AgentLoop {

        // Apply compaction if threshold is configured
        if self.compaction_threshold > 0 {
-            messages = compaction::maybe_compact(messages, self.compaction_threshold);
+            let needs_async =
+                self.compaction_config.use_llm || self.compaction_config.memory_flush_enabled;
+            if needs_async {
+                let outcome = compaction::maybe_compact_with_config(
+                    messages,
+                    self.compaction_threshold,
+                    &self.compaction_config,
+                    &self.agent_id,
+                    &session_id,
+                    Some(&self.driver),
+                    self.growth.as_ref(),
+                )
+                .await;
+                messages = outcome.messages;
+            } else {
+                messages = compaction::maybe_compact(messages, self.compaction_threshold);
+            }
        }

        // Enhance system prompt with growth memories
--- a/desktop/package.json
+++ b/desktop/package.json
@@ -45,7 +45,10 @@
    "lucide-react": "^0.577.0",
    "react": "^19.2.4",
    "react-dom": "^19.2.4",
+    "react-markdown": "^10.1.0",
    "react-window": "^2.2.7",
+    "recharts": "^3.8.1",
+    "remark-gfm": "^4.0.1",
    "smol-toml": "^1.6.1",
    "tailwind-merge": "^3.5.0",
    "tweetnacl": "^1.0.3",
--- a/desktop/pnpm-lock.yaml
+++ b/desktop/pnpm-lock.yaml
--- a/desktop/src-tauri/src/intelligence/compactor.rs
+++ b/desktop/src-tauri/src/intelligence/compactor.rs
@@ -539,14 +539,26 @@ pub fn compactor_check_threshold(

 /// Execute compaction
 #[tauri::command]
-pub fn compactor_compact(
+pub async fn compactor_compact(
    messages: Vec<CompactableMessage>,
    agent_id: String,
    conversation_id: Option<String>,
    config: Option<CompactionConfig>,
 ) -> CompactionResult {
+    let memory_flush = config
+        .as_ref()
+        .map(|c| c.memory_flush_enabled)
+        .unwrap_or(false);
+    let flushed = if memory_flush {
+        flush_old_messages_to_memory(&messages, &agent_id, conversation_id.as_deref()).await
+    } else {
+        0
+    };
+
    let compactor = ContextCompactor::new(config);
-    compactor.compact(&messages, &agent_id, conversation_id.as_deref())
+    let mut result = compactor.compact(&messages, &agent_id, conversation_id.as_deref());
+    result.flushed_memories = flushed;
+    result
 }

 /// Execute compaction with optional LLM-based summary
@@ -558,10 +570,95 @@ pub async fn compactor_compact_llm(
    compaction_config: Option<CompactionConfig>,
    llm_config: Option<LlmSummaryConfig>,
 ) -> CompactionResult {
+    let memory_flush = compaction_config
+        .as_ref()
+        .map(|c| c.memory_flush_enabled)
+        .unwrap_or(false);
+    let flushed = if memory_flush {
+        flush_old_messages_to_memory(&messages, &agent_id, conversation_id.as_deref()).await
+    } else {
+        0
+    };
+
    let compactor = ContextCompactor::new(compaction_config);
-    compactor
+    let mut result = compactor
        .compact_with_llm(&messages, &agent_id, conversation_id.as_deref(), llm_config.as_ref())
-        .await
+        .await;
+    result.flushed_memories = flushed;
+    result
+}
+
+/// Flush important messages from the old (pre-compaction) portion to VikingStorage.
+///
+/// Extracts user messages and key assistant responses as session memories
+/// so that information is preserved even after messages are compacted away.
+async fn flush_old_messages_to_memory(
+    messages: &[CompactableMessage],
+    agent_id: &str,
+    _conversation_id: Option<&str>,
+) -> usize {
+    let storage = match crate::viking_commands::get_storage().await {
+        Ok(s) => s,
+        Err(e) => {
+            tracing::warn!("[Compactor] Cannot get storage for memory flush: {}", e);
+            return 0;
+        }
+    };
+
+    let mut flushed = 0usize;
+    let mut prev_was_user = false;
+
+    for msg in messages {
+        // Flush user messages as session memories (they contain user intent/preferences)
+        if msg.role == "user" && msg.content.len() > 10 {
+            let entry = zclaw_growth::MemoryEntry::new(
+                agent_id,
+                zclaw_growth::MemoryType::Session,
+                "compaction_flush",
+                msg.content.clone(),
+            )
+            .with_importance(4);
+
+            match zclaw_growth::VikingStorage::store(storage.as_ref(), &entry).await {
+                Ok(_) => flushed += 1,
+                Err(e) => {
+                    tracing::debug!("[Compactor] Memory flush failed for user msg: {}", e);
+                }
+            }
+            prev_was_user = true;
+        } else if msg.role == "assistant" && prev_was_user {
+            // Flush the assistant response that follows a user message (contains answers)
+            if msg.content.len() > 20 {
+                let entry = zclaw_growth::MemoryEntry::new(
+                    agent_id,
+                    zclaw_growth::MemoryType::Session,
+                    "compaction_flush",
+                    msg.content.clone(),
+                )
+                .with_importance(3);
+
+                match zclaw_growth::VikingStorage::store(storage.as_ref(), &entry).await {
+                    Ok(_) => flushed += 1,
+                    Err(e) => {
+                        tracing::debug!("[Compactor] Memory flush failed for assistant msg: {}", e);
+                    }
+                }
+            }
+            prev_was_user = false;
+        } else {
+            prev_was_user = false;
+        }
+    }
+
+    if flushed > 0 {
+        tracing::info!(
+            "[Compactor] Flushed {} memories before compaction for agent {}",
+            flushed,
+            agent_id
+        );
+    }
+
+    flushed
 }

 #[cfg(test)]
--- a/desktop/src/components/MessageSearch.tsx
+++ b/desktop/src/components/MessageSearch.tsx
@@ -1,19 +1,26 @@
 import { useState, useEffect, useCallback, useMemo, useRef } from 'react';
 import { motion, AnimatePresence } from 'framer-motion';
-import { Search, X, ChevronUp, ChevronDown, Clock, User, Filter } from 'lucide-react';
+import { Search, X, ChevronUp, ChevronDown, Clock, User, Filter, Globe, MessageSquare } from 'lucide-react';
 import { Button } from './ui';
 import { useChatStore, Message } from '../store/chatStore';
+import { intelligence, PersistentMemory } from '../lib/intelligence-backend';

 export interface SearchFilters {
  sender: 'all' | 'user' | 'assistant';
  timeRange: 'all' | 'today' | 'week' | 'month';
 }

+export type SearchScope = 'session' | 'global';
+
 export interface SearchResult {
  message: Message;
  matchIndices: Array<{ start: number; end: number }>;
 }

+export interface GlobalSearchResult {
+  memory: PersistentMemory;
+}
+
 interface MessageSearchProps {
  onNavigateToMessage: (messageId: string) => void;
 }
@@ -26,6 +33,7 @@ export function MessageSearch({ onNavigateToMessage }: MessageSearchProps) {

  const [isOpen, setIsOpen] = useState(false);
  const [query, setQuery] = useState('');
+  const [scope, setScope] = useState<SearchScope>('session');
  const [filters, setFilters] = useState<SearchFilters>({
    sender: 'all',
    timeRange: 'all',
@@ -33,6 +41,8 @@ export function MessageSearch({ onNavigateToMessage }: MessageSearchProps) {
  const [currentMatchIndex, setCurrentMatchIndex] = useState(0);
  const [showFilters, setShowFilters] = useState(false);
  const [searchHistory, setSearchHistory] = useState<string[]>([]);
+  const [globalResults, setGlobalResults] = useState<GlobalSearchResult[]>([]);
+  const [globalLoading, setGlobalLoading] = useState(false);
  const inputRef = useRef<HTMLInputElement>(null);

  // Load search history from localStorage
@@ -63,6 +73,41 @@ export function MessageSearch({ onNavigateToMessage }: MessageSearchProps) {
    });
  }, []);

+  // Global search: query VikingStorage when scope is 'global'
+  useEffect(() => {
+    if (scope !== 'global' || !query.trim()) {
+      setGlobalResults([]);
+      return;
+    }
+
+    let cancelled = false;
+    const debounceTimer = setTimeout(async () => {
+      setGlobalLoading(true);
+      try {
+        const results = await intelligence.memory.search({
+          query: query.trim(),
+          limit: 20,
+        });
+        if (!cancelled) {
+          setGlobalResults(results.map((memory) => ({ memory })));
+        }
+      } catch (err) {
+        if (!cancelled) {
+          setGlobalResults([]);
+        }
+      } finally {
+        if (!cancelled) {
+          setGlobalLoading(false);
+        }
+      }
+    }, 300);
+
+    return () => {
+      cancelled = true;
+      clearTimeout(debounceTimer);
+    };
+  }, [scope, query]);
+
  // Filter messages by time range
  const filterByTimeRange = useCallback((message: Message, timeRange: SearchFilters['timeRange']): boolean => {
    if (timeRange === 'all') return true;
@@ -245,6 +290,36 @@ export function MessageSearch({ onNavigateToMessage }: MessageSearchProps) {
          >
            <div className="px-4 py-3">
              <form onSubmit={handleSubmit} className="flex items-center gap-2">
+                {/* Scope toggle */}
+                <div className="flex items-center bg-gray-100 dark:bg-gray-700 rounded-lg p-0.5">
+                  <button
+                    type="button"
+                    onClick={() => setScope('session')}
+                    className={`flex items-center gap-1 px-2 py-1 rounded text-xs transition-colors ${
+                      scope === 'session'
+                        ? 'bg-white dark:bg-gray-600 text-orange-600 dark:text-orange-400 shadow-sm'
+                        : 'text-gray-500 dark:text-gray-400 hover:text-gray-600 dark:hover:text-gray-300'
+                    }`}
+                    aria-label="Search current session"
+                  >
+                    <MessageSquare className="w-3 h-3" />
+                    <span className="hidden sm:inline">Session</span>
+                  </button>
+                  <button
+                    type="button"
+                    onClick={() => setScope('global')}
+                    className={`flex items-center gap-1 px-2 py-1 rounded text-xs transition-colors ${
+                      scope === 'global'
+                        ? 'bg-white dark:bg-gray-600 text-orange-600 dark:text-orange-400 shadow-sm'
+                        : 'text-gray-500 dark:text-gray-400 hover:text-gray-600 dark:hover:text-gray-300'
+                    }`}
+                    aria-label="Search all memories"
+                  >
+                    <Globe className="w-3 h-3" />
+                    <span className="hidden sm:inline">Global</span>
+                  </button>
+                </div>
+
                {/* Search input */}
                <div className="flex-1 relative">
                  <Search className="absolute left-3 top-1/2 -translate-y-1/2 w-4 h-4 text-gray-400" />
@@ -253,7 +328,7 @@ export function MessageSearch({ onNavigateToMessage }: MessageSearchProps) {
                    type="text"
                    value={query}
                    onChange={(e) => setQuery(e.target.value)}
-                    placeholder="Search messages..."
+                    placeholder={scope === 'global' ? 'Search all memories...' : 'Search messages...'}
                    className="w-full pl-9 pr-8 py-2 text-sm bg-white dark:bg-gray-800 border border-gray-200 dark:border-gray-700 rounded-lg focus:outline-none focus:ring-2 focus:ring-orange-500 dark:focus:ring-orange-400 focus:border-transparent"
                    aria-label="Search query"
                  />
@@ -269,22 +344,24 @@ export function MessageSearch({ onNavigateToMessage }: MessageSearchProps) {
                  )}
                </div>

-                {/* Filter toggle */}
-                <Button
-                  type="button"
-                  variant={showFilters ? 'secondary' : 'ghost'}
-                  size="sm"
-                  onClick={() => setShowFilters((prev) => !prev)}
-                  className="flex items-center gap-1"
-                  aria-label="Toggle filters"
-                  aria-expanded={showFilters}
-                >
-                  <Filter className="w-4 h-4" />
-                  <span className="hidden sm:inline">Filters</span>
-                </Button>
+                {/* Filter toggle (session only) */}
+                {scope === 'session' && (
+                  <Button
+                    type="button"
+                    variant={showFilters ? 'secondary' : 'ghost'}
+                    size="sm"
+                    onClick={() => setShowFilters((prev) => !prev)}
+                    className="flex items-center gap-1"
+                    aria-label="Toggle filters"
+                    aria-expanded={showFilters}
+                  >
+                    <Filter className="w-4 h-4" />
+                    <span className="hidden sm:inline">Filters</span>
+                  </Button>
+                )}

-                {/* Navigation buttons */}
-                {searchResults.length > 0 && (
+                {/* Navigation buttons (session only) */}
+                {scope === 'session' && searchResults.length > 0 && (
                  <div className="flex items-center gap-1">
                    <span className="text-xs text-gray-500 dark:text-gray-400 px-2">
                      {currentMatchIndex + 1} / {searchResults.length}
@@ -381,8 +458,58 @@ export function MessageSearch({ onNavigateToMessage }: MessageSearchProps) {
                </div>
              )}

-              {/* No results message */}
-              {query && searchResults.length === 0 && (
+              {/* Global search results */}
+              {scope === 'global' && query && (
+                <div className="mt-2 max-h-64 overflow-y-auto">
+                  {globalLoading && (
+                    <div className="text-xs text-gray-500 dark:text-gray-400 text-center py-2">
+                      Searching memories...
+                    </div>
+                  )}
+                  {!globalLoading && globalResults.length === 0 && (
+                    <div className="text-xs text-gray-500 dark:text-gray-400 text-center py-2">
+                      No memories found matching "{query}"
+                    </div>
+                  )}
+                  {!globalLoading && globalResults.length > 0 && (
+                    <div className="space-y-1">
+                      <div className="text-xs text-gray-400 dark:text-gray-500 mb-1">
+                        {globalResults.length} memories found
+                      </div>
+                      {globalResults.map((result) => (
+                        <div
+                          key={result.memory.id}
+                          className="px-2 py-1.5 bg-white dark:bg-gray-700/50 border border-gray-100 dark:border-gray-600 rounded text-xs"
+                        >
+                          <div className="flex items-center gap-1.5 mb-0.5">
+                            <span className="text-orange-500 dark:text-orange-400 font-medium">
+                              {result.memory.memory_type}
+                            </span>
+                            <span className="text-gray-300 dark:text-gray-600">|</span>
+                            <span className="text-gray-400 dark:text-gray-500">
+                              {result.memory.agent_id}
+                            </span>
+                            {result.memory.importance > 5 && (
+                              <span className="text-yellow-500">
+                                {'*'.repeat(Math.min(result.memory.importance - 4, 5))}
+                              </span>
+                            )}
+                          </div>
+                          <div className="text-gray-700 dark:text-gray-300 line-clamp-2">
+                            {highlightSearchMatches(result.memory.content, query)}
+                          </div>
+                          <div className="text-gray-400 dark:text-gray-500 mt-0.5">
+                            {result.memory.created_at.split('T')[0]}
+                          </div>
+                        </div>
+                      ))}
+                    </div>
+                  )}
+                </div>
+              )}
+
+              {/* No results message (session search) */}
+              {scope === 'session' && query && searchResults.length === 0 && (
                <div className="mt-2 text-xs text-gray-500 dark:text-gray-400 text-center py-2">
                  No messages found matching "{query}"
                </div>
--- a/desktop/src/components/ReflectionLog.tsx
+++ b/desktop/src/components/ReflectionLog.tsx
@@ -485,7 +485,7 @@ export function ReflectionLog({
        // Initialize reflection engine with config that allows soul modification
        await intelligenceClient.reflection.init(config);

-        const loadedHistory = await intelligenceClient.reflection.getHistory();
+        const loadedHistory = await intelligenceClient.reflection.getHistory(undefined, agentId);
        setHistory([...loadedHistory].reverse()); // Most recent first

        const proposals = await intelligenceClient.identity.getPendingProposals(agentId);
--- a/desktop/src/components/presentation/PresentationContainer.tsx
+++ b/desktop/src/components/presentation/PresentationContainer.tsx
@@ -18,6 +18,7 @@ import { QuizRenderer } from './renderers/QuizRenderer';

 const SlideshowRenderer = React.lazy(() => import('./renderers/SlideshowRenderer').then(m => ({ default: m.SlideshowRenderer })));
 const DocumentRenderer = React.lazy(() => import('./renderers/DocumentRenderer').then(m => ({ default: m.DocumentRenderer })));
+const ChartRenderer = React.lazy(() => import('./renderers/ChartRenderer').then(m => ({ default: m.ChartRenderer })));

 interface PresentationContainerProps {
  /** Pipeline output data */
@@ -78,7 +79,7 @@ export function PresentationContainer({
    if (supportedTypes && supportedTypes.length > 0) {
      return supportedTypes.filter((t): t is PresentationType => t !== 'auto');
    }
-    return (['quiz', 'slideshow', 'document', 'whiteboard'] as PresentationType[]);
+    return (['quiz', 'slideshow', 'document', 'chart', 'whiteboard'] as PresentationType[]);
  }, [supportedTypes]);

  const renderContent = () => {
@@ -111,11 +112,21 @@ export function PresentationContainer({

      case 'whiteboard':
        return (
-          <div className="flex items-center justify-center h-64 bg-gray-50">
-            <p className="text-gray-500">白板渲染器开发中...</p>
+          <div className="flex flex-col items-center justify-center h-64 bg-gray-50 gap-3">
+            <span className="inline-flex items-center px-3 py-1 rounded-full text-xs font-medium bg-amber-100 text-amber-700">
+              即将推出
+            </span>
+            <p className="text-gray-500">白板渲染器开发中</p>
          </div>
        );

+      case 'chart':
+        return (
+          <React.Suspense fallback={<div className="h-64 animate-pulse bg-gray-100" />}>
+            <ChartRenderer data={data as Parameters<typeof ChartRenderer>[0]['data']} />
+          </React.Suspense>
+        );
+
      default:
        return (
          <div className="flex items-center justify-center h-64 bg-gray-50">
--- a/desktop/src/components/presentation/renderers/ChartRenderer.tsx
+++ b/desktop/src/components/presentation/renderers/ChartRenderer.tsx
@@ -0,0 +1,204 @@
+/**
+ * Chart Renderer
+ *
+ * Renders data as interactive charts using recharts.
+ * Supports: line, bar, pie, scatter, area chart types.
+ */
+
+import {
+  LineChart, Line, BarChart, Bar, PieChart, Pie, Cell,
+  ScatterChart, Scatter, AreaChart, Area,
+  XAxis, YAxis, CartesianGrid, Tooltip, Legend, ResponsiveContainer,
+} from 'recharts';
+import type { ChartData } from '../types';
+
+const DEFAULT_COLORS = [
+  '#3b82f6', '#ef4444', '#22c55e', '#f59e0b', '#8b5cf6',
+  '#ec4899', '#06b6d4', '#f97316', '#14b8a6', '#6366f1',
+];
+
+interface ChartRendererProps {
+  data: ChartData;
+  className?: string;
+}
+
+export function ChartRenderer({ data, className = '' }: ChartRendererProps) {
+  const { type, title, labels, datasets, options } = data;
+
+  // Transform datasets + labels into recharts data format
+  const chartData = (labels || []).map((label, i) => {
+    const point: Record<string, string | number> = { name: label };
+    for (const ds of datasets) {
+      point[ds.label] = ds.data[i] ?? 0;
+    }
+    return point;
+  });
+
+  // If no labels, use index as x-axis
+  if (!labels || labels.length === 0) {
+    const maxLen = Math.max(...datasets.map(ds => ds.data.length), 0);
+    for (let i = 0; i < maxLen; i++) {
+      const point: Record<string, string | number> = { name: `${i + 1}` };
+      for (const ds of datasets) {
+        point[ds.label] = ds.data[i] ?? 0;
+      }
+      chartData.push(point);
+    }
+  }
+
+  const showLegend = options?.plugins?.legend?.display !== false;
+
+  const legendProps = showLegend
+    ? { wrapperStyle: { paddingBottom: 8 } }
+    : undefined;
+
+  const chartTitle = title || options?.plugins?.title?.text;
+
+  const renderChart = () => {
+    switch (type) {
+      case 'line':
+        return (
+          <ResponsiveContainer width="100%" height="100%">
+            <LineChart data={chartData}>
+              <CartesianGrid strokeDasharray="3 3" opacity={0.3} />
+              <XAxis dataKey="name" fontSize={12} />
+              <YAxis fontSize={12} />
+              <Tooltip />
+              {showLegend && <Legend {...legendProps} />}
+              {datasets.map((ds, i) => (
+                <Line
+                  key={ds.label}
+                  type="monotone"
+                  dataKey={ds.label}
+                  stroke={Array.isArray(ds.borderColor) ? ds.borderColor[0] : (ds.borderColor || DEFAULT_COLORS[i % DEFAULT_COLORS.length])}
+                  strokeWidth={2}
+                  dot={{ r: 3 }}
+                  fill={Array.isArray(ds.backgroundColor) ? undefined : (ds.backgroundColor as string | undefined)}
+                />
+              ))}
+            </LineChart>
+          </ResponsiveContainer>
+        );
+
+      case 'bar':
+        return (
+          <ResponsiveContainer width="100%" height="100%">
+            <BarChart data={chartData}>
+              <CartesianGrid strokeDasharray="3 3" opacity={0.3} />
+              <XAxis dataKey="name" fontSize={12} />
+              <YAxis fontSize={12} />
+              <Tooltip />
+              {showLegend && <Legend {...legendProps} />}
+              {datasets.map((ds, i) => (
+                <Bar
+                  key={ds.label}
+                  dataKey={ds.label}
+                  fill={Array.isArray(ds.backgroundColor) ? ds.backgroundColor[0] : (ds.backgroundColor || DEFAULT_COLORS[i % DEFAULT_COLORS.length])}
+                  radius={[4, 4, 0, 0]}
+                />
+              ))}
+            </BarChart>
+          </ResponsiveContainer>
+        );
+
+      case 'pie': {
+        const pieData = datasets.flatMap((ds) =>
+          (labels || ds.data.map((_, i) => `${i + 1}`)).map((label, i) => ({
+            name: label,
+            value: ds.data[i] ?? 0,
+          }))
+        );
+        return (
+          <ResponsiveContainer width="100%" height="100%">
+            <PieChart>
+              <Pie
+                data={pieData}
+                cx="50%"
+                cy="50%"
+                labelLine
+                label={({ name, percent }: { name?: string; percent?: number }) => `${name ?? ''} ${((percent ?? 0) * 100).toFixed(0)}%`}
+                outerRadius="70%"
+                dataKey="value"
+              >
+                {pieData.map((_, i) => (
+                  <Cell key={i} fill={DEFAULT_COLORS[i % DEFAULT_COLORS.length]} />
+                ))}
+              </Pie>
+              <Tooltip />
+              {showLegend && <Legend />}
+            </PieChart>
+          </ResponsiveContainer>
+        );
+      }
+
+      case 'scatter': {
+        const scatterData = datasets.flatMap((ds) =>
+          ds.data.map((val, i) => ({
+            x: labels ? i : i + 1,
+            y: val,
+            name: ds.label,
+          }))
+        );
+        return (
+          <ResponsiveContainer width="100%" height="100%">
+            <ScatterChart>
+              <CartesianGrid strokeDasharray="3 3" opacity={0.3} />
+              <XAxis dataKey="x" name="X" fontSize={12} />
+              <YAxis dataKey="y" name="Y" fontSize={12} />
+              <Tooltip cursor={{ strokeDasharray: '3 3' }} />
+              {showLegend && <Legend />}
+              <Scatter name={datasets[0]?.label || '数据'} data={scatterData} fill={DEFAULT_COLORS[0]} />
+            </ScatterChart>
+          </ResponsiveContainer>
+        );
+      }
+
+      case 'area':
+        return (
+          <ResponsiveContainer width="100%" height="100%">
+            <AreaChart data={chartData}>
+              <CartesianGrid strokeDasharray="3 3" opacity={0.3} />
+              <XAxis dataKey="name" fontSize={12} />
+              <YAxis fontSize={12} />
+              <Tooltip />
+              {showLegend && <Legend {...legendProps} />}
+              {datasets.map((ds, i) => (
+                <Area
+                  key={ds.label}
+                  type="monotone"
+                  dataKey={ds.label}
+                  stroke={Array.isArray(ds.borderColor) ? ds.borderColor[0] : (ds.borderColor || DEFAULT_COLORS[i % DEFAULT_COLORS.length])}
+                  fill={Array.isArray(ds.backgroundColor) ? ds.backgroundColor[0] : (ds.backgroundColor || `${DEFAULT_COLORS[i % DEFAULT_COLORS.length]}33`)}
+                  strokeWidth={2}
+                />
+              ))}
+            </AreaChart>
+          </ResponsiveContainer>
+        );
+
+      default:
+        return <p className="text-gray-500 text-center">不支持的图表类型: {type}</p>;
+    }
+  };
+
+  return (
+    <div className={`flex flex-col h-full ${className}`}>
+      {chartTitle && (
+        <div className="p-4 border-b border-gray-200">
+          <h2 className="text-lg font-semibold text-gray-900">{chartTitle}</h2>
+        </div>
+      )}
+      <div className="flex-1 p-4" style={{ minHeight: 300 }}>
+        {datasets.length === 0 ? (
+          <div className="flex items-center justify-center h-full">
+            <p className="text-gray-400">暂无图表数据</p>
+          </div>
+        ) : (
+          renderChart()
+        )}
+      </div>
+    </div>
+  );
+}
+
+export default ChartRenderer;
--- a/desktop/src/components/presentation/renderers/DocumentRenderer.tsx
+++ b/desktop/src/components/presentation/renderers/DocumentRenderer.tsx
@@ -1,10 +1,13 @@
 /**
 * Document Renderer
 *
- * Renders content as a scrollable document with Markdown support.
+ * Renders content as a scrollable document with full Markdown support
+ * via react-markdown + remark-gfm (tables, strikethrough, task lists, etc.).
 */

 import { useState } from 'react';
+import Markdown from 'react-markdown';
+import remarkGfm from 'remark-gfm';
 import { Download, ExternalLink, Copy } from 'lucide-react';
 import type { DocumentData } from '../types';

@@ -26,12 +29,14 @@ export function DocumentRenderer({

  const handleCopy = async () => {
    try {
-      const textToCopy = typeof data === 'string' ? data : (data.content || JSON.stringify(data, null, 2));
+      const textToCopy = typeof data === 'string'
+        ? data
+        : (data.content || JSON.stringify(data, null, 2));
      await navigator.clipboard.writeText(textToCopy);
      setCopied(true);
      setTimeout(() => setCopied(false), 2000);
-    } catch (error) {
-      console.error('Failed to copy:', error);
+    } catch {
+      // Clipboard API may not be available in all contexts
    }
  };

@@ -46,58 +51,14 @@ export function DocumentRenderer({
    }
  };

-  const renderMarkdown = (content: string): React.ReactNode => {
-    const lines = content.split('\n');
-    const elements: React.ReactNode[] = [];
-
-    for (const line of lines) {
-      const trimmed = line.trim();
-      if (!trimmed) continue;
-
-      if (trimmed.startsWith('# ')) {
-        elements.push(
-          <h1 key={trimmed} className="text-2xl font-bold mb-4">
-            {trimmed.substring(2)}
-          </h1>
-        );
-      } else if (trimmed.startsWith('## ')) {
-        elements.push(
-          <h2 key={trimmed} className="text-xl font-semibold mb-3">
-            {trimmed.substring(3)}
-          </h2>
-        );
-      } else if (trimmed.startsWith('### ')) {
-        elements.push(
-          <h3 key={trimmed} className="text-lg font-medium mb-2">
-            {trimmed.substring(4)}
-          </h3>
-        );
-      } else if (trimmed.startsWith('- ')) {
-        elements.push(
-          <li key={trimmed} className="ml-4 list-disc">
-            {trimmed.substring(2)}
-          </li>
-        );
-      } else if (trimmed.startsWith('```')) {
-        elements.push(
-          <pre key={trimmed} className="bg-gray-900 text-gray-100 p-4 rounded-lg overflow-x-auto text-sm my-2">
-            <code>{trimmed.substring(3, trimmed.length - 3)}</code>
-          </pre>
-        );
-      } else {
-        elements.push(
-          <p key={trimmed} className="mb-2">{trimmed}</p>
-        );
-      }
-    }
-
-    return <div className={className}>{elements}</div>;
-  };
+  const content = typeof data === 'string'
+    ? data
+    : (data.content || JSON.stringify(data, null, 2));

  if (!enableMarkdown) {
    return (
      <div className={`flex flex-col h-full ${className}`}>
-        <pre className="whitespace-pre-wrap text-sm">{JSON.stringify(data, null, 2)}</pre>
+        <pre className="whitespace-pre-wrap text-sm">{content}</pre>
      </div>
    );
  }
@@ -138,10 +99,8 @@ export function DocumentRenderer({
        </div>
      )}

-      <div className="flex-1 overflow-auto p-6">
-        {typeof data === 'string' 
-          ? renderMarkdown(data) 
-          : renderMarkdown(data.content || JSON.stringify(data))}
+      <div className="flex-1 overflow-auto p-6 prose prose-gray max-w-none">
+        <Markdown remarkPlugins={[remarkGfm]}>{content}</Markdown>
      </div>
    </div>
  );
--- a/desktop/src/components/presentation/renderers/SlideshowRenderer.tsx
+++ b/desktop/src/components/presentation/renderers/SlideshowRenderer.tsx
@@ -2,9 +2,12 @@
 * Slideshow Renderer
 *
 * Renders presentation as a slideshow with slide navigation.
+ * Supports: title, content, image, code, twoColumn slide types.
 */

 import { useState, useEffect, useCallback } from 'react';
+import Markdown from 'react-markdown';
+import remarkGfm from 'remark-gfm';
 import {
  ChevronLeft,
  ChevronRight,
@@ -13,7 +16,7 @@ import {
  Play,
  Pause,
 } from 'lucide-react';
-import type { SlideshowData } from '../types';
+import type { SlideshowData, Slide } from '../types';

 interface SlideshowRendererProps {
  data: SlideshowData;
@@ -41,30 +44,6 @@ export function SlideshowRenderer({
  const slides = data.slides || [];
  const totalSlides = slides.length;

-  // Handle keyboard navigation
-  useEffect(() => {
-    const handleKeyDown = (e: KeyboardEvent) => {
-      if (e.key === 'ArrowRight' || e.key === ' ') {
-        handleNext();
-      } else if (e.key === 'ArrowLeft') {
-        handlePrev();
-      } else if (e.key === 'f') {
-        toggleFullscreen();
-      }
-    };
-
-    window.addEventListener('keydown', handleKeyDown);
-    return () => window.removeEventListener('keydown', handleKeyDown);
-  }, []);
-
-  // Auto-play
-  useEffect(() => {
-    if (isPlaying && autoPlayInterval > 0) {
-      const timer = setInterval(handleNext, autoPlayInterval * 1000);
-      return () => clearInterval(timer);
-    }
-  }, [isPlaying, autoPlayInterval]);
-
  const handleNext = useCallback(() => {
    setCurrentIndex((prev) => (prev + 1) % totalSlides);
  }, [totalSlides]);
@@ -77,6 +56,32 @@ export function SlideshowRenderer({
    setIsFullscreen((prev) => !prev);
  }, []);

+  // Handle keyboard navigation
+  useEffect(() => {
+    const handleKeyDown = (e: KeyboardEvent) => {
+      if (e.key === 'ArrowRight' || e.key === ' ') {
+        e.preventDefault();
+        handleNext();
+      } else if (e.key === 'ArrowLeft') {
+        e.preventDefault();
+        handlePrev();
+      } else if (e.key === 'f') {
+        toggleFullscreen();
+      }
+    };
+
+    window.addEventListener('keydown', handleKeyDown);
+    return () => window.removeEventListener('keydown', handleKeyDown);
+  }, [handleNext, handlePrev, toggleFullscreen]);
+
+  // Auto-play
+  useEffect(() => {
+    if (isPlaying && autoPlayInterval > 0) {
+      const timer = setInterval(handleNext, autoPlayInterval * 1000);
+      return () => clearInterval(timer);
+    }
+  }, [isPlaying, autoPlayInterval, handleNext]);
+
  const currentSlide = slides[currentIndex];

  if (!currentSlide) {
@@ -88,26 +93,15 @@ export function SlideshowRenderer({
  }

  return (
-    <div className={`flex flex-col h-full ${isFullscreen ? 'fixed inset-0 z-50 bg-white' : ''} ${className}`}>
+    <div
+      className={`flex flex-col h-full ${
+        isFullscreen ? 'fixed inset-0 z-50 bg-white' : ''
+      } ${className}`}
+    >
      {/* Slide Content */}
      <div className="flex-1 flex items-center justify-center p-8">
        <div className="max-w-4xl w-full">
-          {/* Title */}
-          {currentSlide.title && (
-            <h2 className="text-3xl font-bold text-center mb-6">
-              {currentSlide.title}
-            </h2>
-          )}
-
-          {/* Content rendering would go here */}
-          <div className="text-gray-700">
-            {/* This is simplified - real implementation would render based on content type */}
-            {typeof currentSlide.content === 'string' ? (
-              <p>{currentSlide.content}</p>
-            ) : (
-              <div>Complex content rendering</div>
-            )}
-          </div>
+          <SlideContent slide={currentSlide} />
        </div>
      </div>

@@ -127,7 +121,11 @@ export function SlideshowRenderer({
            disabled={autoPlayInterval === 0}
            className="p-2 hover:bg-gray-200 rounded disabled:opacity-50"
          >
-            {isPlaying ? <Pause className="w-5 h-5" /> : <Play className="w-5 h-5" />}
+            {isPlaying ? (
+              <Pause className="w-5 h-5" />
+            ) : (
+              <Play className="w-5 h-5" />
+            )}
          </button>

          <button
@@ -162,11 +160,124 @@ export function SlideshowRenderer({
      {/* Speaker Notes */}
      {showNotes && currentSlide.notes && (
        <div className="p-4 bg-yellow-50 border-t text-sm text-gray-600">
-          📝 {currentSlide.notes}
+          {currentSlide.notes}
        </div>
      )}
    </div>
  );
 }

+/** Renders a single slide based on its type */
+function SlideContent({ slide }: { slide: Slide }) {
+  switch (slide.type) {
+    case 'title':
+      return (
+        <div className="text-center py-12">
+          {slide.title && (
+            <h1 className="text-4xl font-bold mb-4">{slide.title}</h1>
+          )}
+          {slide.content && (
+            <p className="text-xl text-gray-600 max-w-2xl mx-auto">
+              {slide.content}
+            </p>
+          )}
+        </div>
+      );
+
+    case 'content':
+      return (
+        <div>
+          {slide.title && (
+            <h2 className="text-3xl font-bold text-center mb-6">{slide.title}</h2>
+          )}
+          {slide.content && (
+            <div className="prose prose-gray max-w-none">
+              <Markdown remarkPlugins={[remarkGfm]}>{slide.content}</Markdown>
+            </div>
+          )}
+        </div>
+      );
+
+    case 'image':
+      return (
+        <div className="text-center">
+          {slide.title && (
+            <h2 className="text-2xl font-bold mb-4">{slide.title}</h2>
+          )}
+          {slide.image && (
+            <img
+              src={slide.image}
+              alt={slide.title || '幻灯片图片'}
+              className="max-w-full max-h-[60vh] mx-auto rounded-lg shadow-md"
+            />
+          )}
+          {slide.content && (
+            <p className="mt-4 text-gray-600">{slide.content}</p>
+          )}
+        </div>
+      );
+
+    case 'code':
+      return (
+        <div>
+          {slide.title && (
+            <h2 className="text-2xl font-bold mb-4">{slide.title}</h2>
+          )}
+          {slide.code && (
+            <pre className="bg-gray-900 text-gray-100 p-6 rounded-lg overflow-x-auto text-sm">
+              {slide.language && (
+                <div className="text-xs text-gray-400 mb-3 uppercase tracking-wider">
+                  {slide.language}
+                </div>
+              )}
+              <code>{slide.code}</code>
+            </pre>
+          )}
+          {slide.content && (
+            <p className="mt-4 text-gray-600">{slide.content}</p>
+          )}
+        </div>
+      );
+
+    case 'twoColumn':
+      return (
+        <div>
+          {slide.title && (
+            <h2 className="text-2xl font-bold mb-4">{slide.title}</h2>
+          )}
+          <div className="grid grid-cols-2 gap-8">
+            <div className="prose prose-sm max-w-none">
+              {slide.leftContent && (
+                <Markdown remarkPlugins={[remarkGfm]}>
+                  {slide.leftContent}
+                </Markdown>
+              )}
+            </div>
+            <div className="prose prose-sm max-w-none">
+              {slide.rightContent && (
+                <Markdown remarkPlugins={[remarkGfm]}>
+                  {slide.rightContent}
+                </Markdown>
+              )}
+            </div>
+          </div>
+        </div>
+      );
+
+    default:
+      return (
+        <div>
+          {slide.title && (
+            <h2 className="text-2xl font-bold mb-4">{slide.title}</h2>
+          )}
+          {slide.content && (
+            <div className="prose prose-gray max-w-none">
+              <Markdown remarkPlugins={[remarkGfm]}>{slide.content}</Markdown>
+            </div>
+          )}
+        </div>
+      );
+  }
+}
+
 export default SlideshowRenderer;
--- a/desktop/src/lib/intelligence-backend.ts
+++ b/desktop/src/lib/intelligence-backend.ts
@@ -347,8 +347,8 @@ export const reflection = {
    return invoke('reflection_reflect', { agentId, memories });
  },

-  async getHistory(limit?: number): Promise<ReflectionResult[]> {
-    return invoke('reflection_get_history', { limit });
+  async getHistory(limit?: number, agentId?: string): Promise<ReflectionResult[]> {
+    return invoke('reflection_get_history', { limit, agentId });
  },

  async getState(): Promise<ReflectionState> {
--- a/desktop/src/lib/intelligence-client.ts
+++ b/desktop/src/lib/intelligence-client.ts
@@ -691,7 +691,7 @@ const fallbackReflection = {
    return result;
  },

-  async getHistory(limit?: number): Promise<ReflectionResult[]> {
+  async getHistory(limit?: number, _agentId?: string): Promise<ReflectionResult[]> {
    const l = limit ?? 10;
    return fallbackReflection._history.slice(-l).reverse();
  },
@@ -1318,13 +1318,13 @@ export const intelligenceClient = {
      return fallbackReflection.reflect(agentId, memories);
    },

-    getHistory: async (limit?: number): Promise<ReflectionResult[]> => {
+    getHistory: async (limit?: number, agentId?: string): Promise<ReflectionResult[]> => {
      if (isTauriRuntime()) {
        return tauriInvoke('reflection.getHistory', () =>
-          intelligence.reflection.getHistory(limit)
+          intelligence.reflection.getHistory(limit, agentId)
        );
      }
-      return fallbackReflection.getHistory(limit);
+      return fallbackReflection.getHistory(limit, agentId);
    },

    getState: async (): Promise<ReflectionState> => {
--- a/docs/features/DEEP_AUDIT_REPORT.md
+++ b/docs/features/DEEP_AUDIT_REPORT.md
@@ -522,6 +522,7 @@ PipelinesPanel.tsx → workflowStore.runPipeline()
 | **P2** | M5 | 自主授权集成到执行链路 | 1-2d | 下周 | ✅ 已修复 |
 | **P2** | M3 | hand_approve 使用 hand_name 参数 | 1h | 下周 | ✅ 已修复 |
 | **P2** | L2 | 清理 gatewayStore 废弃引用 | 1h | 下周 | ✅ 已确认（仅注释） |
+| **P2** | S9 | 消息搜索仅当前会话 — 新增 Global 模式 | 1d | 下周 | ✅ 已修复 |
 | **P3** | M6 | 实现语义路由 | 2-3d | 下个迭代 |
 | **P3** | L1 | Pipeline 并行执行 | 2d | 下个迭代 |
 | **P3** | L3 | Wasm/Native 技能模式 | 3-5d | 长期 |
@@ -570,6 +571,7 @@ ZCLAW 的核心架构（通信、状态管理、安全认证、聊天、Agent
 13. ~~**反思历史只存单条**~~ ✅ 已修复 — 累积存储到 reflection:history 数组
 14. ~~**身份回滚 UI 缺失**~~ ✅ 已实现 — IdentityChangeProposal.tsx HistoryItem
 15. **28 处 dead_code 标注**中大部分是合理的预留功能，少数是遗留代码
-16. **剩余 P2/P3 项**: 反思 LLM 分析、跨会话搜索、语义路由、Pipeline 并行等
+16. **剩余 P2/P3 项**: 反思 LLM 分析、语义路由、Pipeline 并行等
+17. ~~**消息搜索仅当前会话**~~ ✅ 已修复 — MessageSearch 新增 Global 模式，调用 VikingStorage memory_search 跨会话搜索记忆

-**累计修复 22 项** (P0×3 + P1×8 + P2×6 + 误判×2 + 审计×3)，系统真实可用率从 ~50% 提升到 ~80%。剩余 P3 项为增强功能，不阻塞核心使用。
+**累计修复 23 项** (P0×3 + P1×8 + P2×7 + 误判×2 + 审计×3)，系统真实可用率从 ~50% 提升到 ~80%。剩余 P3 项为增强功能，不阻塞核心使用。
--- a/docs/features/README.md
+++ b/docs/features/README.md
@@ -1,12 +1,12 @@
 # ZCLAW 功能全景文档

-> **版本**: v0.6.3
+> **版本**: v0.6.4
 > **更新日期**: 2026-03-27
 > **项目状态**: 完整 Rust Workspace 架构，10 个核心 Crates，69 技能，Pipeline DSL + Smart Presentation + Agent Growth System
-> **整体完成度**: ~62% (基于 2026-03-27 深度审计 + 两轮修复后)
+> **整体完成度**: ~72% (基于 2026-03-27 深度审计 + 四轮修复后)
 > **架构**: Tauri 桌面应用，Rust Workspace (10 crates) + React 前端
 >
-> **审计修复 (2026-03-27)**: 修复 2 个 CRITICAL + 6 个 HIGH + 2 个 MEDIUM 问题，详见 [DEEP_AUDIT_REPORT.md](./DEEP_AUDIT_REPORT.md)
+> **审计修复 (2026-03-27)**: 累计修复 23 项 (P0×3 + P1×8 + P2×7 + 误判×2 + 审计×3)，详见 [DEEP_AUDIT_REPORT.md](./DEEP_AUDIT_REPORT.md)

 > **重要**: ZCLAW 采用 Rust Workspace 架构，包含 10 个分层 Crates (types → memory → runtime → kernel → skills/hands/protocols/pipeline/growth/channels)，所有核心能力集成在 Tauri 桌面应用中

@@ -26,7 +26,7 @@

 | 文档 | 功能 | 成熟度 | 测试覆盖 |
 |------|------|--------|---------|
-| [00-chat-interface.md](01-core-features/00-chat-interface.md) | 聊天界面 | L3-L4 (90%) | 高 |
+| [00-chat-interface.md](01-core-features/00-chat-interface.md) | 聊天界面 | L4 (92%) | 高 |
 | [01-agent-clones.md](01-core-features/01-agent-clones.md) | Agent 分身 | L3 (85%) | 高 |
 | [02-hands-system.md](01-core-features/02-hands-system.md) | Hands 系统 | L3 (60%) | 中 |
 | 工作流引擎 | 工作流引擎 | L3 (80%) | 中 |
@@ -144,7 +144,7 @@
 | S6 | 导出功能清理 (PPTX/PDF 友好提示) | P2 | ✅ 完成 |
 | S7 | Compactor 接入聊天流程 | P1 | ✅ 完成 |
 | S8 | 定时任务 KernelClient 支持 | P1 | 待开始 |
-| S9 | 添加消息搜索功能 | P1 | 待开始 |
+| S9 | 添加消息搜索功能 | P1 | ✅ 完成 (Session + Global 双模式) |
 | S10 | 优化错误提示 | P1 | 待开始 |

 ### 2.2 中期计划 (1-2 月)
@@ -289,6 +289,7 @@ skills  hands  protocols  pipeline  growth  channels

 | 日期 | 版本 | 变更内容 |
 |------|------|---------|
+| 2026-03-27 | v0.6.4 | **审计修复 (P2 第四轮)**: S9 消息搜索跨会话 (Session + Global 双模式，VikingStorage 搜索)、M5-补 自主授权后端守卫、M3 hand_approve 参数修复、M4-补 反思历史累积存储、心跳历史持久化。累计修复 23 项，整体完成度 65%→72%。|
 | 2026-03-27 | v0.6.3 | **审计修复 (P1/P2)**: H3 记忆双存储统一到 VikingStorage、H4 心跳引擎持久化 + 启动恢复、M4 反思结果持久化。整体完成度 58%→62%。|
 | 2026-03-27 | v0.6.2 | **审计修复 (P0/P1)**: C1 PromptOnly LLM 集成、C2 反思引擎空记忆修复、H7 Agent Store 接口适配、H8 Hand 审批检查、M1 幽灵命令注册、H1/H2 demo 标记、H5 归档过时报告。整体完成度 50%→58%。|
 | 2026-03-27 | v0.6.1 | **功能完整性修复**: 激活 LoopGuard 循环防护、实现 CapabilityManager.validate() 安全验证、handStore/workflowStore KernelClient 适配器、Credits 标注开发中、Skills 动态化、ScheduledTasks localStorage 降级、token 用量追踪 |
@@ -332,7 +333,18 @@ skills  hands  protocols  pipeline  growth  channels
 | 反思结果持久化 | M4 | `reflect()` 后持久化 ReflectionState/Result 到 VikingStorage，重启后自动恢复 |
 | 清理 dead_code warnings | — | PersistentMemoryStore impl 添加 `#[allow(dead_code)]`，移除未使用的 `build_uri` |

-### 7.3 代码清理
+### 7.3 审计修复 (P2 第四轮)
+
+| 修复项 | ID | 说明 |
+|--------|-----|------|
+| 自主授权后端守卫 | M5-补 | `hand_execute`/`skill_execute` 接收 `autonomy_level` 参数，三级守卫 (supervised/assisted/autonomous) |
+| hand_approve 参数 | M3 | 移除 `_` 前缀，添加审计日志，返回值包含 hand_name |
+| 反思历史累积 | M4-补 | 新增 `reflection:history:{agent_id}` 数组(最多 20 条)，向后兼容 `reflection:latest` |
+| 心跳历史持久化 | — | `tick()` 同步存储历史到 VikingStorage，`heartbeat_init()` 恢复历史 |
+| 身份回滚 UI | — | 确认 `IdentityChangeProposal.tsx` 已实现 HistoryItem + restoreSnapshot |
+| 跨会话消息搜索 | S9 | MessageSearch 新增 Session/Global 双模式，Global 调用 `memory_search` 搜索 VikingStorage |
+
+### 7.4 代码清理

 | 清理项 | 说明 |
 |--------|------|
@@ -352,4 +364,4 @@ skills  hands  protocols  pipeline  growth  channels
 | ScheduledTasks 持久化 | 添加 localStorage 降级，刷新不丢失 |
 | Token 用量追踪 | chatStore 新增 addTokenUsage/getTotalTokens |

-> **审计说明**: 成熟度等级已根据代码审计调整为实际值。Identity Evolution 标注为 L2 (70%) 是因为其 `dead_code` 属性属于 Tauri 运行时模式（在 Tauri 上下文中实际被调用），而非真正的死代码。Reflection Engine L2 (65%) 因核心反思逻辑尚未深度迭代。
+> **审计说明**: 成熟度等级已根据代码审计调整为实际值。Identity Evolution 标注为 L2 (70%) 是因为其 `dead_code` 属性属于 Tauri 运行时模式（在 Tauri 上下文中实际被调用），而非真正的死代码。Reflection Engine L2 (65%) 因核心反思逻辑尚未深度迭代。累计修复 23 项后整体完成度从 ~50% 提升到 ~72%。