fix(agent): 12 项 agent 对话链路全栈修复
Some checks are pending
CI / Lint & TypeCheck (push) Waiting to run
CI / Unit Tests (push) Waiting to run
CI / Build Frontend (push) Waiting to run
CI / Rust Check (push) Waiting to run
CI / Security Scan (push) Waiting to run
CI / E2E Tests (push) Blocked by required conditions
Some checks are pending
CI / Lint & TypeCheck (push) Waiting to run
CI / Unit Tests (push) Waiting to run
CI / Build Frontend (push) Waiting to run
CI / Rust Check (push) Waiting to run
CI / Security Scan (push) Waiting to run
CI / E2E Tests (push) Blocked by required conditions
深端到端验证发现 12 个问题,6 Phase 全栈修复: Phase 5 — 快速 UX 修复: - #9: SimpleSidebar 添加新对话按钮 (SquarePen + useChatStore) - #5: 模型列表 JOIN provider_keys 过滤无 API Key 的模型 - #11: AgentOnboardingWizard 焦点领域增加 4 行业选项 (医疗健康/教育培训/金融财务/法律合规) Phase 1 — ButlerPanel 记忆修复: - #2a: MemorySection URI 从 viking://agent/.../memories/ 修正为 agent://.../ - #2b: "立即分析对话"按钮现在触发 extractAndStoreMemories Phase 2 — FTS5 中文分词: - #4: FTS5 tokenizer 从 unicode61 切换到 trigram,原生支持 CJK - 自动迁移:检测旧 unicode61 表并重建索引 - sanitize_fts_query 支持中文引号短语查询 Phase 3 — 跨会话身份持久化: - #6-8: 重新启用 USER.md 注入系统提示词 (截断前 10 行) Phase 4 — Agent 面板同步: - #1,#10: listClones 从 4 字段扩展到完整映射 (soul/userProfile 解析 nickname/emoji/userName/userRole) - updateClone 通过 identity 系统同步 nickname→SOUL.md 和 userName/userRole→USER.md Phase 6 — Agent 创建容错: - #12: createFromTemplate 增加 SaaS 不可用 fallback 验证: tsc --noEmit ✅ cargo check ✅
This commit is contained in:
@@ -132,13 +132,16 @@ impl SqliteStorage {
|
||||
.map_err(|e| ZclawError::StorageError(format!("Failed to create memories table: {}", e)))?;
|
||||
|
||||
// Create FTS5 virtual table for full-text search
|
||||
// Use trigram tokenizer for CJK (Chinese/Japanese/Korean) support.
|
||||
// unicode61 cannot tokenize CJK characters, causing memory search to fail.
|
||||
// trigram indexes overlapping 3-character slices, works well for all languages.
|
||||
sqlx::query(
|
||||
r#"
|
||||
CREATE VIRTUAL TABLE IF NOT EXISTS memories_fts USING fts5(
|
||||
uri,
|
||||
content,
|
||||
keywords,
|
||||
tokenize='unicode61'
|
||||
tokenize='trigram'
|
||||
)
|
||||
"#,
|
||||
)
|
||||
@@ -189,6 +192,46 @@ impl SqliteStorage {
|
||||
.await
|
||||
.map_err(|e| ZclawError::StorageError(format!("Failed to create metadata table: {}", e)))?;
|
||||
|
||||
// Migration: Rebuild FTS5 table if using old unicode61 tokenizer (can't handle CJK)
|
||||
// Check tokenizer by inspecting the existing FTS5 table definition
|
||||
let needs_rebuild: bool = sqlx::query_scalar::<_, i64>(
|
||||
"SELECT COUNT(*) FROM sqlite_master WHERE type='table' AND name='memories_fts' AND sql LIKE '%unicode61%'"
|
||||
)
|
||||
.fetch_one(&self.pool)
|
||||
.await
|
||||
.unwrap_or(0) > 0;
|
||||
|
||||
if needs_rebuild {
|
||||
tracing::info!("[SqliteStorage] Rebuilding FTS5 table: unicode61 → trigram for CJK support");
|
||||
// Drop old FTS5 table
|
||||
let _ = sqlx::query("DROP TABLE IF EXISTS memories_fts")
|
||||
.execute(&self.pool)
|
||||
.await;
|
||||
// Recreate with trigram tokenizer
|
||||
sqlx::query(
|
||||
r#"
|
||||
CREATE VIRTUAL TABLE IF NOT EXISTS memories_fts USING fts5(
|
||||
uri,
|
||||
content,
|
||||
keywords,
|
||||
tokenize='trigram'
|
||||
)
|
||||
"#,
|
||||
)
|
||||
.execute(&self.pool)
|
||||
.await
|
||||
.map_err(|e| ZclawError::StorageError(format!("Failed to recreate FTS5 table: {}", e)))?;
|
||||
// Reindex all existing memories into FTS5
|
||||
let reindexed = sqlx::query(
|
||||
"INSERT INTO memories_fts (uri, content, keywords) SELECT uri, content, keywords FROM memories"
|
||||
)
|
||||
.execute(&self.pool)
|
||||
.await
|
||||
.map(|r| r.rows_affected())
|
||||
.unwrap_or(0);
|
||||
tracing::info!("[SqliteStorage] FTS5 rebuild complete, reindexed {} entries", reindexed);
|
||||
}
|
||||
|
||||
tracing::info!("[SqliteStorage] Database schema initialized");
|
||||
Ok(())
|
||||
}
|
||||
@@ -378,19 +421,37 @@ impl SqliteStorage {
|
||||
/// Strips these and keeps only alphanumeric + CJK tokens with length > 1,
|
||||
/// then joins them with `OR` for broad matching.
|
||||
fn sanitize_fts_query(query: &str) -> String {
|
||||
let terms: Vec<String> = query
|
||||
.to_lowercase()
|
||||
.split(|c: char| !c.is_alphanumeric())
|
||||
.filter(|s| !s.is_empty() && s.len() > 1)
|
||||
.map(|s| s.to_string())
|
||||
.collect();
|
||||
// trigram tokenizer requires quoted phrases for substring matching
|
||||
// and needs at least 3 characters per term to produce results.
|
||||
let lower = query.to_lowercase();
|
||||
|
||||
if terms.is_empty() {
|
||||
return String::new();
|
||||
// Check if query contains CJK characters — trigram handles them natively
|
||||
let has_cjk = lower.chars().any(|c| {
|
||||
matches!(c, '\u{4E00}'..='\u{9FFF}' | '\u{3400}'..='\u{4DBF}' | '\u{F900}'..='\u{FAFF}')
|
||||
});
|
||||
|
||||
if has_cjk {
|
||||
// For CJK, use the full query as a quoted phrase for substring matching
|
||||
// trigram will match any 3-char subsequence
|
||||
if lower.len() >= 3 {
|
||||
format!("\"{}\"", lower)
|
||||
} else {
|
||||
String::new()
|
||||
}
|
||||
} else {
|
||||
// For non-CJK, split into terms and join with OR
|
||||
let terms: Vec<String> = lower
|
||||
.split(|c: char| !c.is_alphanumeric())
|
||||
.filter(|s| !s.is_empty() && s.len() > 1)
|
||||
.map(|s| format!("\"{}\"", s))
|
||||
.collect();
|
||||
|
||||
if terms.is_empty() {
|
||||
return String::new();
|
||||
}
|
||||
|
||||
terms.join(" OR ")
|
||||
}
|
||||
|
||||
// Join with OR so any term can match (broad recall, then rerank by similarity)
|
||||
terms.join(" OR ")
|
||||
}
|
||||
|
||||
/// Fetch memories by scope with importance-based ordering.
|
||||
|
||||
Reference in New Issue
Block a user