diff --git a/desktop/src-tauri/src/intelligence/mod.rs b/desktop/src-tauri/src/intelligence/mod.rs
index 32adcd7..55a6e8d 100644
--- a/desktop/src-tauri/src/intelligence/mod.rs
+++ b/desktop/src-tauri/src/intelligence/mod.rs
@@ -32,6 +32,8 @@ pub mod reflection;
 pub mod identity;
 pub mod validation;
 pub mod extraction_adapter;
+pub mod pain_aggregator;
+pub mod solution_generator;
 
 // Re-export main types for convenience
 pub use heartbeat::HeartbeatEngineState;
diff --git a/desktop/src-tauri/src/intelligence/pain_aggregator.rs b/desktop/src-tauri/src/intelligence/pain_aggregator.rs
new file mode 100644
index 0000000..6b870fd
--- /dev/null
+++ b/desktop/src-tauri/src/intelligence/pain_aggregator.rs
@@ -0,0 +1,570 @@
+//! Pain Point Aggregator — detect, merge, and score recurring user difficulties.
+//!
+//! Collects pain signals from conversations, merges similar ones with increasing
+//! confidence, and surfaces high-confidence pain points for solution generation.
+
+use std::sync::Arc;
+
+use chrono::{DateTime, Utc};
+use serde::{Deserialize, Serialize};
+use tokio::sync::RwLock;
+use uuid::Uuid;
+use zclaw_types::Result;
+
+// ---------------------------------------------------------------------------
+// Data structures
+// ---------------------------------------------------------------------------
+
+/// A recurring difficulty or systemic problem the user encounters.
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct PainPoint {
+    pub id: String,
+    pub agent_id: String,
+    pub user_id: String,
+    pub summary: String,
+    pub category: String,
+    pub severity: PainSeverity,
+    pub evidence: Vec<PainEvidence>,
+    pub occurrence_count: u32,
+    pub first_seen: DateTime<Utc>,
+    pub last_seen: DateTime<Utc>,
+    pub confidence: f64,
+    pub status: PainStatus,
+}
+
+/// A single piece of evidence supporting a pain point.
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct PainEvidence {
+    pub when: DateTime<Utc>,
+    pub user_said: String,
+    pub why_flagged: String,
+}
+
+/// How severe the pain point is.
+#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize, Default)]
+#[serde(rename_all = "lowercase")]
+pub enum PainSeverity {
+    #[default]
+    Low,
+    Medium,
+    High,
+}
+
+/// Lifecycle status of a pain point.
+#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
+#[serde(rename_all = "lowercase")]
+pub enum PainStatus {
+    Detected,
+    Confirmed,
+    Solving,
+    Solved,
+    Dismissed,
+}
+
+impl PainPoint {
+    /// Create a new pain point with a single evidence entry.
+    pub fn new(
+        agent_id: &str,
+        user_id: &str,
+        summary: &str,
+        category: &str,
+        severity: PainSeverity,
+        user_said: &str,
+        why_flagged: &str,
+    ) -> Self {
+        let now = Utc::now();
+        Self {
+            id: Uuid::new_v4().to_string(),
+            agent_id: agent_id.to_string(),
+            user_id: user_id.to_string(),
+            summary: summary.to_string(),
+            category: category.to_string(),
+            severity,
+            evidence: vec![PainEvidence {
+                when: now,
+                user_said: user_said.to_string(),
+                why_flagged: why_flagged.to_string(),
+            }],
+            occurrence_count: 1,
+            first_seen: now,
+            last_seen: now,
+            confidence: 0.25,
+            status: PainStatus::Detected,
+        }
+    }
+}
+
+// ---------------------------------------------------------------------------
+// LLM analysis result
+// ---------------------------------------------------------------------------
+
+/// Result of LLM-based pain point analysis on a conversation.
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct PainAnalysisResult {
+    pub found: bool,
+    #[serde(default)]
+    pub summary: String,
+    #[serde(default)]
+    pub category: String,
+    #[serde(default)]
+    pub severity: PainSeverity,
+    #[serde(default)]
+    pub evidence: String,
+}
+
+// ---------------------------------------------------------------------------
+// PainAggregator — cross-session merge + confidence scoring
+// ---------------------------------------------------------------------------
+
+/// Aggregates pain points across conversations, merging similar ones
+/// and escalating confidence as evidence accumulates.
+pub struct PainAggregator {
+    pain_points: Arc<RwLock<Vec<PainPoint>>>,
+}
+
+impl PainAggregator {
+    pub fn new() -> Self {
+        Self {
+            pain_points: Arc::new(RwLock::new(Vec::new())),
+        }
+    }
+
+    /// Merge a new pain point with an existing similar one, or create a new entry.
+    pub async fn merge_or_create(&self, new_pain: PainPoint) -> Result<PainPoint> {
+        let mut points = self.pain_points.write().await;
+
+        let similar_idx = points.iter().position(|p| {
+            p.agent_id == new_pain.agent_id
+                && p.category == new_pain.category
+                && Self::summaries_similar(&p.summary, &new_pain.summary)
+        });
+
+        if let Some(idx) = similar_idx {
+            let existing = &mut points[idx];
+            existing.evidence.extend(new_pain.evidence);
+            existing.occurrence_count += 1;
+            existing.last_seen = Utc::now();
+            existing.confidence = Self::calculate_confidence(existing);
+            if new_pain.severity as u8 > existing.severity as u8 {
+                existing.severity = new_pain.severity;
+            }
+            if existing.occurrence_count >= 2 && existing.status == PainStatus::Detected {
+                existing.status = PainStatus::Confirmed;
+            }
+            Ok(existing.clone())
+        } else {
+            let result = new_pain.clone();
+            points.push(new_pain);
+            Ok(result)
+        }
+    }
+
+    /// Get all high-confidence pain points for an agent.
+    pub async fn get_high_confidence_pains(
+        &self,
+        agent_id: &str,
+        min_confidence: f64,
+    ) -> Vec<PainPoint> {
+        let points = self.pain_points.read().await;
+        points
+            .iter()
+            .filter(|p| {
+                p.agent_id == agent_id
+                    && p.confidence >= min_confidence
+                    && p.status != PainStatus::Solved
+                    && p.status != PainStatus::Dismissed
+            })
+            .cloned()
+            .collect()
+    }
+
+    /// List all pain points for an agent.
+    pub async fn list_pain_points(&self, agent_id: &str) -> Vec<PainPoint> {
+        let points = self.pain_points.read().await;
+        points
+            .iter()
+            .filter(|p| p.agent_id == agent_id)
+            .cloned()
+            .collect()
+    }
+
+    /// List all pain points across all agents (for internal use).
+    pub async fn list_all_pain_points(&self) -> Vec<PainPoint> {
+        let points = self.pain_points.read().await;
+        points.clone()
+    }
+
+    /// Update the status of a pain point.
+    pub async fn update_status(&self, pain_id: &str, status: PainStatus) -> Result<()> {
+        let mut points = self.pain_points.write().await;
+        if let Some(p) = points.iter_mut().find(|p| p.id == pain_id) {
+            p.status = status;
+        }
+        Ok(())
+    }
+
+    // -- internal helpers ----------------------------------------------------
+
+    fn summaries_similar(a: &str, b: &str) -> bool {
+        let a_lower = a.to_lowercase();
+        let b_lower = b.to_lowercase();
+
+        if a_lower.contains(&b_lower) || b_lower.contains(&a_lower) {
+            return true;
+        }
+
+        let a_ngrams = Self::char_ngrams(&a_lower, 2);
+        let b_ngrams = Self::char_ngrams(&b_lower, 2);
+        if a_ngrams.is_empty() || b_ngrams.is_empty() {
+            return false;
+        }
+        let shared = a_ngrams.iter().filter(|ng| b_ngrams.contains(ng)).count();
+        let overlap_ratio = shared as f64 / a_ngrams.len().min(b_ngrams.len()) as f64;
+        overlap_ratio >= 0.4
+    }
+
+    fn char_ngrams(text: &str, n: usize) -> Vec<String> {
+        let chars: Vec<char> = text.chars().collect();
+        if chars.len() < n {
+            return Vec::new();
+        }
+        chars.windows(n).map(|w| w.iter().collect()).collect()
+    }
+
+    fn calculate_confidence(pain: &PainPoint) -> f64 {
+        let base = (pain.occurrence_count as f64 * 0.25).min(0.75);
+        let days_since_last = (Utc::now() - pain.last_seen).num_days().max(0) as f64;
+        let decay = 1.0 - (days_since_last / 90.0);
+        let severity_boost = match pain.severity {
+            PainSeverity::High => 0.1,
+            PainSeverity::Medium => 0.05,
+            PainSeverity::Low => 0.0,
+        };
+        (base * decay.max(0.0) + severity_boost).min(1.0)
+    }
+}
+
+impl Default for PainAggregator {
+    fn default() -> Self {
+        Self::new()
+    }
+}
+
+// ---------------------------------------------------------------------------
+// Rule-based pain signal detection
+// ---------------------------------------------------------------------------
+
+const FRUSTRATION_SIGNALS: &[&str] = &[
+    "烦死了", "又来了", "每次都", "受不了", "还是不行",
+    "又出错", "第三次", "第四次", "第五次", "怎么又",
+    "老是", "一直", "还是没解决", "退了", "被退",
+    "又要改", "搞不定", "没办法",
+];
+
+/// Analyze messages for pain signals using rule-based detection.
+pub fn analyze_for_pain_signals(messages: &[zclaw_types::Message]) -> Option<PainAnalysisResult> {
+    let mut user_frustrations: Vec<String> = Vec::new();
+
+    for msg in messages {
+        if let zclaw_types::Message::User { content } = msg {
+            for signal in FRUSTRATION_SIGNALS {
+                if content.contains(signal) {
+                    user_frustrations.push(content.clone());
+                    break;
+                }
+            }
+        }
+    }
+
+    if user_frustrations.is_empty() {
+        return None;
+    }
+
+    let combined = user_frustrations.join(" ");
+    let severity = if combined.contains("烦死了")
+        || combined.contains("受不了")
+        || user_frustrations.len() >= 3
+    {
+        PainSeverity::High
+    } else if user_frustrations.len() >= 2 {
+        PainSeverity::Medium
+    } else {
+        PainSeverity::Low
+    };
+
+    let summary = extract_summary(&user_frustrations[0]);
+    let category = classify_category(&combined);
+
+    Some(PainAnalysisResult {
+        found: true,
+        summary,
+        category,
+        severity,
+        evidence: user_frustrations.join("; "),
+    })
+}
+
+fn extract_summary(text: &str) -> String {
+    let trimmed = text.trim();
+    if trimmed.chars().count() <= 50 {
+        trimmed.to_string()
+    } else {
+        let end = trimmed.char_indices().take_while(|(i, _)| *i < 50).last();
+        match end {
+            Some((i, c)) => trimmed[..i + c.len_utf8()].to_string() + "…",
+            None => trimmed.chars().take(50).collect::<String>() + "…",
+        }
+    }
+}
+
+fn classify_category(text: &str) -> String {
+    let lower = text.to_lowercase();
+    if ["包", "货", "出口", "退"].iter().any(|k| lower.contains(k)) {
+        "logistics".to_string()
+    } else if ["合规", "法规", "标准"].iter().any(|k| lower.contains(k)) {
+        "compliance".to_string()
+    } else if ["客户", "投诉", "反馈"].iter().any(|k| lower.contains(k)) {
+        "customer".to_string()
+    } else if ["报价", "价格", "成本"].iter().any(|k| lower.contains(k)) {
+        "pricing".to_string()
+    } else if ["系统", "软件", "电脑"].iter().any(|k| lower.contains(k)) {
+        "technology".to_string()
+    } else {
+        "general".to_string()
+    }
+}
+
+// ---------------------------------------------------------------------------
+// Tauri commands
+// ---------------------------------------------------------------------------
+
+use std::sync::OnceLock;
+
+use super::solution_generator::{Proposal, ProposalStatus, SolutionGenerator};
+
+static PAIN_AGGREGATOR: OnceLock<Arc<PainAggregator>> = OnceLock::new();
+static SOLUTION_GENERATOR: OnceLock<Arc<SolutionGenerator>> = OnceLock::new();
+
+fn pain_store() -> Arc<PainAggregator> {
+    PAIN_AGGREGATOR.get_or_init(|| Arc::new(PainAggregator::new())).clone()
+}
+
+fn solution_store() -> Arc<SolutionGenerator> {
+    SOLUTION_GENERATOR.get_or_init(|| Arc::new(SolutionGenerator::new())).clone()
+}
+
+/// List all pain points for an agent.
+#[tauri::command]
+pub async fn butler_list_pain_points(agent_id: String) -> std::result::Result<Vec<PainPoint>, String> {
+    Ok(pain_store().list_pain_points(&agent_id).await)
+}
+
+/// Record a new pain point from conversation analysis.
+#[tauri::command]
+pub async fn butler_record_pain_point(
+    agent_id: String,
+    user_id: String,
+    summary: String,
+    category: String,
+    severity: String,
+    user_said: String,
+    why_flagged: String,
+) -> std::result::Result<PainPoint, String> {
+    let sev = match severity.as_str() {
+        "high" => PainSeverity::High,
+        "medium" => PainSeverity::Medium,
+        _ => PainSeverity::Low,
+    };
+    let pain = PainPoint::new(&agent_id, &user_id, &summary, &category, sev, &user_said, &why_flagged);
+    pain_store().merge_or_create(pain).await.map_err(|e| e.to_string())
+}
+
+/// List all proposals for an agent.
+#[tauri::command]
+pub async fn butler_list_proposals(agent_id: String) -> std::result::Result<Vec<Proposal>, String> {
+    let store = pain_store();
+    let gen = solution_store();
+    let pains = store.list_pain_points(&agent_id).await;
+    let mut all_proposals = Vec::new();
+    for pain in &pains {
+        let proposals = gen.list_proposals(&[pain.id.clone()]).await;
+        all_proposals.extend(proposals);
+    }
+    Ok(all_proposals)
+}
+
+/// Generate a solution for a high-confidence pain point.
+#[tauri::command]
+pub async fn butler_generate_solution(pain_id: String) -> std::result::Result<Proposal, String> {
+    let store = pain_store();
+    let gen = solution_store();
+    let all_pains = store.list_all_pain_points().await;
+    let pain = all_pains
+        .into_iter()
+        .find(|p| p.id == pain_id)
+        .ok_or_else(|| format!("Pain point {} not found", pain_id))?;
+
+    if pain.confidence < 0.7 {
+        return Err("Pain point confidence too low to generate solution".into());
+    }
+
+    let proposal = gen.generate_solution(&pain).await;
+    store
+        .update_status(&pain_id, PainStatus::Solving)
+        .await
+        .map_err(|e| e.to_string())?;
+    Ok(proposal)
+}
+
+/// Update the status of a proposal.
+#[tauri::command]
+pub async fn butler_update_proposal_status(
+    proposal_id: String,
+    status: String,
+) -> std::result::Result<(), String> {
+    let new_status = match status.as_str() {
+        "accepted" => ProposalStatus::Accepted,
+        "rejected" => ProposalStatus::Rejected,
+        "completed" => ProposalStatus::Completed,
+        _ => ProposalStatus::Pending,
+    };
+    solution_store()
+        .update_status(&proposal_id, new_status)
+        .await
+        .ok_or_else(|| format!("Proposal {} not found", proposal_id))?;
+    Ok(())
+}
+
+// ---------------------------------------------------------------------------
+// Tests
+// ---------------------------------------------------------------------------
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use zclaw_types::Message;
+
+    #[test]
+    fn test_pain_point_serialization() {
+        let pp = PainPoint::new(
+            "agent-1", "user-1", "出口包装合规反复出错",
+            "compliance", PainSeverity::High, "又要改包装", "用户第三次提到包装问题",
+        );
+        assert_eq!(pp.summary, "出口包装合规反复出错");
+        assert_eq!(pp.occurrence_count, 1);
+        assert_eq!(pp.status, PainStatus::Detected);
+
+        let json = serde_json::to_string(&pp).unwrap();
+        let decoded: PainPoint = serde_json::from_str(&json).unwrap();
+        assert_eq!(decoded.summary, "出口包装合规反复出错");
+        assert_eq!(decoded.severity, PainSeverity::High);
+    }
+
+    #[test]
+    fn test_merge_similar_pain_points() {
+        let aggregator = PainAggregator::new();
+        let rt = tokio::runtime::Runtime::new().unwrap();
+
+        let p1 = PainPoint::new("agent-1", "user-1", "出口包装不合格",
+            "compliance", PainSeverity::Medium, "包装被退了", "用户首次提到");
+        let merged1 = rt.block_on(aggregator.merge_or_create(p1)).unwrap();
+        assert_eq!(merged1.occurrence_count, 1);
+
+        let p2 = PainPoint::new("agent-1", "user-1", "包装不合格又要改",
+            "compliance", PainSeverity::High, "又要改包装", "用户第二次提到");
+        let merged2 = rt.block_on(aggregator.merge_or_create(p2)).unwrap();
+        assert_eq!(merged2.occurrence_count, 2);
+        assert!(merged2.confidence > 0.3);
+        assert_eq!(merged2.status, PainStatus::Confirmed);
+    }
+
+    #[test]
+    fn test_different_categories_not_merged() {
+        let aggregator = PainAggregator::new();
+        let rt = tokio::runtime::Runtime::new().unwrap();
+
+        let p1 = PainPoint::new("agent-1", "user-1", "出口包装不合格",
+            "compliance", PainSeverity::Medium, "包装被退", "包装问题");
+        let p2 = PainPoint::new("agent-1", "user-1", "客户投诉服务态度",
+            "customer", PainSeverity::High, "客户又投诉了", "客户问题");
+
+        rt.block_on(aggregator.merge_or_create(p1)).unwrap();
+        let merged2 = rt.block_on(aggregator.merge_or_create(p2)).unwrap();
+        assert_eq!(merged2.occurrence_count, 1);
+    }
+
+    #[test]
+    fn test_confidence_calculation() {
+        let mut pain = PainPoint::new("agent-1", "user-1", "test",
+            "general", PainSeverity::High, "test", "test");
+        pain.occurrence_count = 3;
+        pain.last_seen = Utc::now();
+
+        let conf = PainAggregator::calculate_confidence(&pain);
+        assert!(conf > 0.5, "3 occurrences + High severity should be > 0.5, got {}", conf);
+        assert!(conf <= 1.0);
+    }
+
+    #[test]
+    fn test_analyze_frustration_signals() {
+        let messages = vec![
+            Message::user("这批货又被退了"),
+            Message::assistant("很抱歉听到这个消息"),
+            Message::user("又要改包装，烦死了，第三次了"),
+        ];
+        let result = analyze_for_pain_signals(&messages);
+        assert!(result.is_some());
+        let analysis = result.unwrap();
+        assert!(analysis.found);
+        assert_eq!(analysis.severity, PainSeverity::High);
+        assert!(!analysis.summary.is_empty());
+    }
+
+    #[test]
+    fn test_analyze_no_frustration() {
+        let messages = vec![
+            Message::user("今天天气不错"),
+            Message::assistant("是的，天气很好"),
+        ];
+        assert!(analyze_for_pain_signals(&messages).is_none());
+    }
+
+    #[test]
+    fn test_classify_category() {
+        assert_eq!(classify_category("出口包装被退了"), "logistics");
+        assert_eq!(classify_category("合规标准变了"), "compliance");
+        assert_eq!(classify_category("客户投诉很多"), "customer");
+        assert_eq!(classify_category("报价太低了"), "pricing");
+        assert_eq!(classify_category("系统又崩了"), "technology");
+        assert_eq!(classify_category("随便聊聊天"), "general");
+    }
+
+    #[test]
+    fn test_severity_ordering() {
+        let messages = vec![
+            Message::user("这又来了"),
+            Message::user("还是不行"),
+        ];
+        let result = analyze_for_pain_signals(&messages);
+        assert!(result.is_some());
+        assert_eq!(result.unwrap().severity, PainSeverity::Medium);
+    }
+
+    #[test]
+    fn test_get_high_confidence_filters() {
+        let aggregator = PainAggregator::new();
+        let rt = tokio::runtime::Runtime::new().unwrap();
+
+        for i in 0..3 {
+            let p = PainPoint::new("agent-1", "user-1", "包装不合格反复出错",
+                "compliance", PainSeverity::High,
+                &format!("第{}次出现", i + 1), "recurring");
+            rt.block_on(aggregator.merge_or_create(p)).unwrap();
+        }
+
+        let high = rt.block_on(aggregator.get_high_confidence_pains("agent-1", 0.5));
+        assert!(!high.is_empty());
+        assert!(high[0].confidence >= 0.5);
+        assert_eq!(high[0].occurrence_count, 3);
+    }
+}
diff --git a/desktop/src-tauri/src/intelligence/solution_generator.rs b/desktop/src-tauri/src/intelligence/solution_generator.rs
new file mode 100644
index 0000000..fee734d
--- /dev/null
+++ b/desktop/src-tauri/src/intelligence/solution_generator.rs
@@ -0,0 +1,343 @@
+//! Solution Generator — transforms high-confidence pain points into actionable proposals.
+//!
+//! When a PainPoint reaches confidence >= 0.7, the generator creates a Proposal
+//! with concrete steps derived from available skills and pipeline templates.
+
+use chrono::{DateTime, Utc};
+use serde::{Deserialize, Serialize};
+use uuid::Uuid;
+
+use super::pain_aggregator::{PainEvidence, PainPoint, PainSeverity};
+
+// ---------------------------------------------------------------------------
+// Proposal data structures
+// ---------------------------------------------------------------------------
+
+/// A proposed solution for a confirmed pain point.
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct Proposal {
+    pub id: String,
+    pub pain_point_id: String,
+    pub title: String,
+    pub description: String,
+    pub steps: Vec<ProposalStep>,
+    pub status: ProposalStatus,
+    pub evidence_chain: Vec<PainEvidence>,
+    pub confidence_at_creation: f64,
+    pub created_at: DateTime<Utc>,
+    pub updated_at: DateTime<Utc>,
+}
+
+/// A single step in a proposal.
+#[derive(Debug, Clone, Serialize, Deserialize)]
+pub struct ProposalStep {
+    pub index: u32,
+    pub action: String,
+    pub detail: String,
+    pub skill_hint: Option<String>,
+}
+
+/// Lifecycle status of a proposal.
+#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
+#[serde(rename_all = "lowercase")]
+pub enum ProposalStatus {
+    Pending,
+    Accepted,
+    Rejected,
+    Completed,
+}
+
+impl Proposal {
+    /// Generate a proposal from a confirmed pain point.
+    pub fn from_pain_point(pain: &PainPoint) -> Self {
+        let title = format!("解决: {}", pain.summary);
+        let description = format!(
+            "检测到用户在「{}」方面遇到反复困难（出现{}次，置信度{:.0}%）。\n\
+             证据：{}",
+            pain.category,
+            pain.occurrence_count,
+            pain.confidence * 100.0,
+            pain.evidence
+                .iter()
+                .map(|e| e.user_said.as_str())
+                .collect::<Vec<_>>()
+                .join("；")
+        );
+
+        let steps = Self::generate_steps(pain);
+
+        Self {
+            id: Uuid::new_v4().to_string(),
+            pain_point_id: pain.id.clone(),
+            title,
+            description,
+            steps,
+            status: ProposalStatus::Pending,
+            evidence_chain: pain.evidence.clone(),
+            confidence_at_creation: pain.confidence,
+            created_at: Utc::now(),
+            updated_at: Utc::now(),
+        }
+    }
+
+    /// Generate actionable steps based on pain category and severity.
+    fn generate_steps(pain: &PainPoint) -> Vec<ProposalStep> {
+        let mut steps = Vec::new();
+
+        match pain.category.as_str() {
+            "logistics" => {
+                steps.push(ProposalStep {
+                    index: 1,
+                    action: "梳理流程".into(),
+                    detail: "整理当前物流/包装流程，标注易出错环节".into(),
+                    skill_hint: Some("researcher".into()),
+                });
+                steps.push(ProposalStep {
+                    index: 2,
+                    action: "制定检查清单".into(),
+                    detail: "为易出错环节创建标准检查清单".into(),
+                    skill_hint: Some("collector".into()),
+                });
+                if pain.severity == PainSeverity::High {
+                    steps.push(ProposalStep {
+                        index: 3,
+                        action: "建立预警机制".into(),
+                        detail: "设置自动检查，提前拦截常见错误".into(),
+                        skill_hint: Some("researcher".into()),
+                    });
+                }
+            }
+            "compliance" => {
+                steps.push(ProposalStep {
+                    index: 1,
+                    action: "法规梳理".into(),
+                    detail: "汇总当前适用的法规要求".into(),
+                    skill_hint: Some("researcher".into()),
+                });
+                steps.push(ProposalStep {
+                    index: 2,
+                    action: "差距分析".into(),
+                    detail: "对比当前实践与法规要求的差距".into(),
+                    skill_hint: Some("researcher".into()),
+                });
+            }
+            "customer" => {
+                steps.push(ProposalStep {
+                    index: 1,
+                    action: "问题分类".into(),
+                    detail: "将客户投诉按类型和频率分类".into(),
+                    skill_hint: Some("collector".into()),
+                });
+                steps.push(ProposalStep {
+                    index: 2,
+                    action: "改进方案".into(),
+                    detail: "针对高频问题提出改进建议".into(),
+                    skill_hint: Some("researcher".into()),
+                });
+            }
+            "pricing" => {
+                steps.push(ProposalStep {
+                    index: 1,
+                    action: "成本分析".into(),
+                    detail: "分析当前报价与成本的匹配度".into(),
+                    skill_hint: Some("researcher".into()),
+                });
+            }
+            "technology" => {
+                steps.push(ProposalStep {
+                    index: 1,
+                    action: "问题诊断".into(),
+                    detail: "记录系统问题的具体表现和触发条件".into(),
+                    skill_hint: Some("researcher".into()),
+                });
+                steps.push(ProposalStep {
+                    index: 2,
+                    action: "替代方案".into(),
+                    detail: "评估是否有更好的工具或流程替代".into(),
+                    skill_hint: Some("researcher".into()),
+                });
+            }
+            _ => {
+                steps.push(ProposalStep {
+                    index: 1,
+                    action: "深入了解".into(),
+                    detail: format!("深入了解「{}」问题的具体细节", pain.summary),
+                    skill_hint: None,
+                });
+            }
+        }
+
+        steps
+    }
+}
+
+// ---------------------------------------------------------------------------
+// SolutionGenerator — manages proposals lifecycle
+// ---------------------------------------------------------------------------
+
+use std::sync::Arc;
+use tokio::sync::RwLock;
+
+/// Manages proposal generation from confirmed pain points.
+pub struct SolutionGenerator {
+    proposals: Arc<RwLock<Vec<Proposal>>>,
+}
+
+impl SolutionGenerator {
+    pub fn new() -> Self {
+        Self {
+            proposals: Arc::new(RwLock::new(Vec::new())),
+        }
+    }
+
+    /// Generate a proposal for a high-confidence pain point.
+    pub async fn generate_solution(&self, pain: &PainPoint) -> Proposal {
+        let proposal = Proposal::from_pain_point(pain);
+        let mut proposals = self.proposals.write().await;
+        proposals.push(proposal.clone());
+        proposal
+    }
+
+    /// List all proposals for a given agent's pain points.
+    pub async fn list_proposals(&self, pain_point_ids: &[String]) -> Vec<Proposal> {
+        let proposals = self.proposals.read().await;
+        proposals
+            .iter()
+            .filter(|p| pain_point_ids.contains(&p.pain_point_id))
+            .cloned()
+            .collect()
+    }
+
+    /// Update the status of a proposal.
+    pub async fn update_status(&self, proposal_id: &str, status: ProposalStatus) -> Option<Proposal> {
+        let mut proposals = self.proposals.write().await;
+        if let Some(p) = proposals.iter_mut().find(|p| p.id == proposal_id) {
+            p.status = status;
+            p.updated_at = Utc::now();
+            Some(p.clone())
+        } else {
+            None
+        }
+    }
+}
+
+impl Default for SolutionGenerator {
+    fn default() -> Self {
+        Self::new()
+    }
+}
+
+// ---------------------------------------------------------------------------
+// Tests
+// ---------------------------------------------------------------------------
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use crate::intelligence::pain_aggregator::PainStatus;
+
+    #[test]
+    fn test_proposal_from_pain_point() {
+        let pain = PainPoint::new(
+            "agent-1",
+            "user-1",
+            "出口包装不合格被退回",
+            "logistics",
+            PainSeverity::High,
+            "又退了",
+            "recurring issue",
+        );
+        let proposal = Proposal::from_pain_point(&pain);
+
+        assert!(!proposal.id.is_empty());
+        assert_eq!(proposal.pain_point_id, pain.id);
+        assert!(proposal.title.contains("出口包装"));
+        assert!(!proposal.steps.is_empty());
+        assert_eq!(proposal.status, ProposalStatus::Pending);
+        assert_eq!(proposal.evidence_chain.len(), 1);
+    }
+
+    #[test]
+    fn test_logistics_generates_checklist() {
+        let pain = PainPoint::new(
+            "agent-1",
+            "user-1",
+            "包装问题",
+            "logistics",
+            PainSeverity::Medium,
+            "test",
+            "test",
+        );
+        let proposal = Proposal::from_pain_point(&pain);
+        assert!(proposal.steps.len() >= 2);
+        assert!(proposal.steps[0].skill_hint.is_some());
+    }
+
+    #[test]
+    fn test_high_severity_gets_extra_step() {
+        let pain_low = PainPoint::new(
+            "agent-1",
+            "user-1",
+            "test",
+            "logistics",
+            PainSeverity::Low,
+            "test",
+            "test",
+        );
+        let pain_high = PainPoint::new(
+            "agent-1",
+            "user-1",
+            "test",
+            "logistics",
+            PainSeverity::High,
+            "test",
+            "test",
+        );
+
+        let p_low = Proposal::from_pain_point(&pain_low);
+        let p_high = Proposal::from_pain_point(&pain_high);
+        assert!(p_high.steps.len() > p_low.steps.len());
+    }
+
+    #[test]
+    fn test_proposal_serialization() {
+        let pain = PainPoint::new(
+            "agent-1",
+            "user-1",
+            "test",
+            "general",
+            PainSeverity::Low,
+            "test",
+            "test",
+        );
+        let proposal = Proposal::from_pain_point(&pain);
+        let json = serde_json::to_string(&proposal).unwrap();
+        let decoded: Proposal = serde_json::from_str(&json).unwrap();
+        assert_eq!(decoded.id, proposal.id);
+        assert_eq!(decoded.status, ProposalStatus::Pending);
+    }
+
+    #[tokio::test]
+    async fn test_solution_generator_lifecycle() {
+        let gen = SolutionGenerator::new();
+
+        let pain = PainPoint::new(
+            "agent-1",
+            "user-1",
+            "test",
+            "general",
+            PainSeverity::Medium,
+            "test",
+            "test",
+        );
+        let proposal = gen.generate_solution(&pain).await;
+        assert_eq!(proposal.status, ProposalStatus::Pending);
+
+        let proposals = gen.list_proposals(&[pain.id.clone()]).await;
+        assert_eq!(proposals.len(), 1);
+
+        let updated = gen.update_status(&proposal.id, ProposalStatus::Accepted).await;
+        assert!(updated.is_some());
+        assert_eq!(updated.unwrap().status, ProposalStatus::Accepted);
+    }
+}
diff --git a/desktop/src-tauri/src/lib.rs b/desktop/src-tauri/src/lib.rs
index a861c45..9cd287b 100644
--- a/desktop/src-tauri/src/lib.rs
+++ b/desktop/src-tauri/src/lib.rs
@@ -355,6 +355,12 @@ pub fn run() {
             kernel_commands::mcp::mcp_stop_service,
             kernel_commands::mcp::mcp_list_services,
             kernel_commands::mcp::mcp_call_tool,
+            // Butler commands
+            intelligence::pain_aggregator::butler_list_pain_points,
+            intelligence::pain_aggregator::butler_record_pain_point,
+            intelligence::pain_aggregator::butler_generate_solution,
+            intelligence::pain_aggregator::butler_list_proposals,
+            intelligence::pain_aggregator::butler_update_proposal_status,
         ])
         .run(tauri::generate_context!())
         .expect("error while running tauri application");
diff --git a/desktop/src/lib/viking-client.ts b/desktop/src/lib/viking-client.ts
index edbe098..858fe04 100644
--- a/desktop/src/lib/viking-client.ts
+++ b/desktop/src/lib/viking-client.ts
@@ -202,18 +202,8 @@ export async function extractAndStoreMemories(
 }
 
 /**
- * Inject relevant memories into prompt for enhanced context
- */
-export async function injectVikingPrompt(
-  agentId: string,
-  basePrompt: string,
-  userInput: string,
-  maxTokens?: number
-): Promise<string> {
-  return invoke<string>('viking_inject_prompt', {
-    agentId,
-    basePrompt,
-    userInput,
-    maxTokens,
-  });
+ * Get butler insights data - pain points and proposals for the but agentId }
+  return invoke<ButlerPainPoint[]>('butler_list_pain_points', { agentId });
 }
+
+