chore: checkpoint release follow-up work

2026-10-16 15:34:08 +08:00 · 2026-06-08 01:54:08 +08:00
parent 8803d564a0
commit 05e7df9a87
9 changed files with 36 additions and 539 deletions
@@ -247,12 +247,17 @@
 - `internal/roadmap/lime-aster-codex-alignment-roadmap.md`
 - `lime-rs/src/commands/agent_cmd.rs::agent_generate_title`
 - `lime-rs/src/commands/persona_cmd.rs::generate_persona`
- `lime-rs/src/commands/theme_context_cmd.rs::aster_agent_theme_context_search`

 这份历史档案与专用命令仍可保留各自职责，但不再承担 Query Loop 唯一事实源职责。
-这三条命令属于专用一次性会话能力：允许显式拼自己的临时 `SessionConfig`，但不能参与 submit turn、runtime queue 或 evidence 真相定义。
+这两条命令属于专用一次性会话能力：允许显式拼自己的临时 `SessionConfig`，但不能参与 submit turn、runtime queue 或 evidence 真相定义。
 它们允许为本地 auxiliary session 附带最小 `lime_runtime` metadata，用于记录 `task_profile / routing_decision / cost_state` 一类辅助任务分类事实；必要时也可以把该 auxiliary session 的 `execution_runtime` 诊断快照回传到命令结果，但这份快照只服务该一次性会话自己的诊断与可观测性，不进入 current Query Loop 的 thread / turn 真相。
-当前命令层允许保留的原始执行面只剩这 4 处：`action_runtime` 属于 current 恢复链，`agent_generate_title`、`persona_cmd` 与 `theme_context_cmd` 属于受控 compat 一次性命令。
+当前命令层允许保留的原始执行面只剩这 3 处：`action_runtime` 属于 current 恢复链，`agent_generate_title` 与 `persona_cmd` 属于受控 compat 一次性命令。旧 `aster_agent_theme_context_search` 已下线，主题上下文搜索 current 事实源是 `src/lib/api/themeContextSearch.ts -> App Server agentSession/start + agentSession/turn/start + agentSession/read`。
+
+### `dead`
+
+- `lime-rs/src/commands/theme_context_cmd.rs::aster_agent_theme_context_search`
+
+旧主题上下文搜索 Tauri 命令不得重新注册到 legacy desktop facade；如果后续继续增强主题上下文搜索，只能扩展 App Server current session turn 网关与 RuntimeCore/backend。

 ### `deprecated`

@@ -1601,8 +1601,9 @@ git diff --check
 585. P3.274 live WebSearch/WebFetch 输出证据门禁：只读子代理指出工具 `status=completed` 即算通过仍可能弱化“真实 WebFetch 打开页面并确认事实”的证据。本轮只认领 `scripts/lib/claw-chat-live-web-tool-evidence.mjs`、`scripts/lib/claw-chat-live-web-tool-evidence.test.mjs`、`scripts/claw-chat-ready-streaming-smoke.mjs`、`scripts/lib/live-provider-smoke-gate.test.mjs` 与本计划 / 矩阵记录；helper 现在为 required tools 计算 `outputPresent`、`allRequiredOutputPresent`、`allRequiredOutputPresentForTurn`，行为测试覆盖完成态但空输出的 `WebFetch` 不满足 output present，live smoke verdict 新增 `liveWebRequiredToolOutputsPresent`。该刀不运行真实 Provider；完整整体目标仍需显式授权后跑联网 E2E。
 586. P3.275 Electron release/updater 领域脚本继续收缩：继续沿 Forge-only release/updater 与 `scripts/` 根目录治理目标推进，本轮只认领 Electron release/updater helper、release workflow guard、scripts governance baseline、package / quality / docs 引用与本计划记录，不碰 Rust/App Server/Agent Chat 并行写集。已把 `prepare-github-release-assets.mjs`、`plan-r2-release-cleanup.mjs`、`release-updater-manifest.test.mjs` 迁入 `scripts/electron/prepare-github-release-assets.mjs`、`scripts/electron/r2-release-cleanup-plan.mjs`、`scripts/electron/release-updater-assets.test.mjs`；`release-updater-manifest.mjs` 已按 `dead` 删除，不再作为 current 或 compat 入口。`.github/workflows/release.yml`、`package.json`、Electron current guards、contract / quality guards、`scripts/README.md` 与 `internal/roadmap/appserver/release-updater.md` 均已同步新路径；旧 `scripts/(prepare-github-release-assets|plan-r2-release-cleanup|release-updater-manifest)` current 路径扫描无命中。`scripts/` 根目录基线从 P3.270 的 `rootFiles=105` 继续收缩为 `rootFiles=94 retired=0 untrackedNew=0`。
 587. P3.275 验证 / 边界：修复 `scripts/electron/release-workflow-guard.mjs` 中 R2 cleanup step 名称与 `.github/workflows/release.yml` 的 `Clean old updater assets from Cloudflare R2` 对齐后，定向验证通过：`npm run governance:scripts` 输出 `rootFiles=94 retired=0 untrackedNew=0`；`npm run governance:electron-release-workflow` 通过；`npx vitest run "scripts/electron/release-updater-assets.test.mjs" "scripts/electron/release-workflow-guard.test.mjs" "scripts/electron/current-docs-guard.test.mjs" --silent=passed-only --disableConsoleIntercept` 通过 3 文件 23 项；`node --check` 覆盖 release/updater 迁移脚本与 release workflow guard 通过。该刀只治理 release/updater 脚本事实源和回流守卫，不声称 Windows Squirrel runner、macOS DMG/sign/notarize GitHub Actions 或 live Provider WebSearch/WebFetch E2E 已完成。
-588. P3.276 scripts 根目录 release bucket 清零：继续按用户“一次性治理掉旧的”要求收掉 P3.275 后剩余的唯一 release bucket 根入口。本轮把 `scripts/knowledge-release-scope-report.mjs` 迁入 `scripts/knowledge/release-scope-report.mjs`，`package.json#knowledge:release-scope-report`、`scripts/README.md`、Knowledge PRD / completion audit / exec plan 引用同步到领域目录；脚本自身的 Knowledge 范围匹配补充 `scripts/knowledge/`，避免迁移后把自己的领域脚本误判为 unknown。`scripts/script-root-governance-baseline.json` 删除旧根入口，`npm run governance:scripts` 输出 `rootFiles=93 retired=0 untrackedNew=0`，根目录 bucket 只剩 `governance=4 / misc=28 / quality=31 / smoke=30`，`release` bucket 已清零。旧 `scripts/knowledge-release-scope-report.mjs` 归类为 `dead` 路径，只保留历史 evidence JSON 中的旧审计记录，不作为 current 入口。
 588. P3.276 Claw live smoke 证据边界显式化：继续沿 live Provider 授权前防误判主线推进，本轮只认领 `scripts/claw-chat-ready-streaming-smoke.mjs`、`scripts/lib/live-provider-smoke-gate.test.mjs`、本计划与前端矩阵记录，不触碰 `lime-rs/crates/app-server/src/local_data_source.rs`、`runtime.rs`、`processor.rs`、`runtime_backend.rs`、`electron/appServerHost.ts` 或 Agent Chat runtime 写集。`scripts/claw-chat-ready-streaming-smoke.mjs` 的 summary 新增 `e2eBoundary`，固定 runner 为 `playwright-chromium`、browser automation 为 `chromium.launchPersistentContext`、current bridge 为 `DevBridge transport=electron-host`，证明路径写明为 `Chromium GUI -> DevBridge -> Electron Desktop Host IPC -> App Server JSON-RPC -> RuntimeCore/backend`，并显式标注 `electronRendererIpcE2e=false` / `electronLaunch=false`；`scripts/lib/live-provider-smoke-gate.test.mjs` 同步静态守卫这些字段，防止后续把 DevBridge(electron-host transport) live smoke 误称为 Playwright `_electron.launch()` renderer IPC E2E。该刀不运行真实 Provider、不新增 App Server method、不新增 mock fallback；真实 Electron renderer IPC 证据仍应由独立 Electron smoke / fixture 承接。
 589. P3.276 验证 / 边界：本刀最低验证目标是 `node --check "scripts/claw-chat-ready-streaming-smoke.mjs"`、`npx vitest run "scripts/lib/live-provider-smoke-gate.test.mjs" --silent=passed-only --disableConsoleIntercept`、相关文件 Prettier / `git diff --check` 与未授权 gate `npm run smoke:claw-chat-ready-streaming -- --timeout-ms 30000`。完整整体目标仍未完成：必须在用户明确授权真实模型 / 联网 Provider 后运行 `npm run smoke:claw-chat-ready-streaming -- --allow-live-provider --timeout-ms 180000`，并按真实失败修复 provider、WebSearch/WebFetch tool runtime、streaming event 和 `agentSession/event -> agentSession/read` 回流问题；不能用 mock、external fixture 或 DevBridge 口径描述替代 live E2E。
 590. P3.277 App Server cancel 语义对齐 Codex：针对“等这么长时间对么 / 是否走偏”，复核 `/Users/coso/Documents/dev/rust/codex/codex-rs/core/src/tasks/mod.rs` 后确认 Codex interrupt 只给运行任务约 `100ms` 优雅退出窗口，随后以 `TurnAborted` 作为一等终态；App Server 侧不能把 `agentSession/turn/cancel` JSON-RPC 响应绑在 external backend / live provider cleanup 上。本轮只认领 `lime-rs/crates/app-server/src/runtime.rs`、`scripts/agent-runtime/claw-chat-current-fixture-smoke.mjs`、`.codex/skills/lime-product-e2e-loop/references/product-e2e-loop.md` 与本计划记录，不碰 `lime-rs/crates/app-server/src/local_data_source.rs`，不新增生产 mock fallback，不重新接 Tauri / legacy command。`RuntimeCore::cancel_turn(...)` 现在先通过 current read model 写入 `turn.canceled` 并返回 `AgentSessionTurnCancelResponse`，再后台 best-effort 调用 backend `cancel_turn(...)`；迟到的 `message.delta` / `turn.final_done` / `turn.completed` 不再覆盖已进入 `Completed / Failed / Canceled` 的终态 turn，避免“点停止后又变 completed / 又继续输出”。这保持事实源仍在 App Server JSON-RPC / RuntimeCore，Electron 只负责 IPC / sidecar，backend cancel 仍必须 eventually 可观测。
 591. P3.277 验证 / Electron fixture：新增 Rust 回归 `cancel_turn_returns_canceled_without_waiting_for_backend_cancel` 和 `canceled_turn_ignores_late_runtime_events`，构造 backend cancel 永不返回与 cancel 后迟到事件，证明 App Server current read model 能快速进入 `canceled` 且不被迟到完成事件污染。验证通过：`cargo test --manifest-path "lime-rs/Cargo.toml" -p app-server cancel_turn --lib`，2 项；`cargo test --manifest-path "lime-rs/Cargo.toml" -p app-server canceled_turn_ignores_late_runtime_events --lib`，1 项；`cargo test --manifest-path "lime-rs/Cargo.toml" -p app-server final_done_runtime_event_marks_turn_completed --lib`；`cargo test --manifest-path "lime-rs/Cargo.toml" -p app-server append_external_runtime_events_keeps_sequence_and_turn_scope --lib`；`cargo test --manifest-path "lime-rs/Cargo.toml" -p app-server json_lines_loop_streams_external_backend_events_before_turn_response --lib`。前端 / 契约验证通过：`npx vitest run "src/components/agent/chat/hooks/agentStreamFlowControl.test.ts" "src/lib/api/agentRuntime/threadClient.test.ts" --silent=passed-only --disableConsoleIntercept`，2 文件 27 项；`node "scripts/check-app-server-client-contract.mjs"`，207 项。真实 Electron fixture：`npm run smoke:claw-chat-current-fixture -- --scenario cancel --timeout-ms 180000` 约 6 秒通过，summary session `claw-chat-current-1780854193018-58393`，证明 GUI 点击停止、`agentSession/turn/cancel`、read model `canceled`、输入框恢复和 backend ledger `turnCancel` eventually 到达；随后 `npm run smoke:claw-chat-current-fixture -- --timeout-ms 180000` 约 5 秒通过，summary session `claw-chat-current-1780854208226-59427`，证明“整理今天的国际新闻”完成态仍走 current App Server JSON-RPC + external fixture backend，且不调用正式模型后端。剩余完整目标仍是显式授权后的 live Provider streaming / WebSearch-WebFetch 实际联网 E2E。
+592. P3.278 scripts 根目录 release bucket 清零：继续按用户“一次性治理掉旧的”要求收掉 P3.275 后剩余的唯一 release bucket 根入口。本轮把 `scripts/knowledge-release-scope-report.mjs` 迁入 `scripts/knowledge/release-scope-report.mjs`，`package.json#knowledge:release-scope-report`、`scripts/README.md`、Knowledge PRD / completion audit / exec plan 引用同步到领域目录；脚本自身的 Knowledge 范围匹配补充 `scripts/knowledge/`，避免迁移后把自己的领域脚本误判为 unknown。`scripts/script-root-governance-baseline.json` 删除旧根入口，`npm run governance:scripts` 输出 `rootFiles=93 retired=0 untrackedNew=0`，根目录 bucket 只剩 `governance=4 / misc=28 / quality=31 / smoke=30`，`release` bucket 已清零。旧 `scripts/knowledge-release-scope-report.mjs` 归类为 `dead` 路径，只保留历史 evidence JSON 中的旧审计记录，不作为 current 入口。
+593. P3.279 Codex cancel 语义复核 / 不再长等：收到“等这么长时间对么 / 参考 codex / 已经走偏”的反馈后，本轮只读对照 `/Users/coso/Documents/dev/rust/codex/codex-rs/core/src/tasks/mod.rs` 与 `app-server/src/bespoke_event_handling.rs`，确认 Codex interrupt 路径通过 cancellation token 取消运行任务，只给 `GRACEFULL_INTERRUPTION_TIMEOUT_MS=100` 的短暂优雅窗口，然后 abort task 并发出 `TurnAborted`；App Server 侧收到 `TurnAborted` 会回复 pending interrupt 并把 turn 投影为 interrupted/completed 状态。这证明 Lime 不能用长 timeout、等待 backend cleanup 或合成 `final_done` 作为停止交付标准。复核当前 Lime 主链后确认：`RuntimeCore::cancel_turn(...)` 已先写入 `turn.canceled` read model，再后台 best-effort 调 backend cancel；前端 `stopActiveAgentStream(...)` 已先解除 active stream、恢复输入框并异步 cancel + refresh；`threadClient` 会把 cancel response notifications 路由回当前 stream event。复测通过：`cargo test --manifest-path "lime-rs/Cargo.toml" -p app-server cancel_turn --lib`，2 项；`cargo test --manifest-path "lime-rs/Cargo.toml" -p app-server canceled_turn_ignores_late_runtime_events --lib`，1 项；`npx vitest run "src/components/agent/chat/hooks/agentStreamFlowControl.test.ts" "src/lib/api/agentRuntime/threadClient.test.ts" --silent=passed-only --disableConsoleIntercept`，2 文件 27 项；`node "scripts/check-app-server-client-contract.mjs"`，207 项；`npm run smoke:claw-chat-current-fixture -- --scenario cancel --timeout-ms 180000` 约 5.7 秒通过，session `claw-chat-current-1780854725258-94095`；`npm run smoke:claw-chat-current-fixture -- --timeout-ms 180000` 约 4.4 秒通过，session `claw-chat-current-1780854745753-95800`，证明“整理今天的国际新闻”在 Electron GUI + App Server JSON-RPC + external fixture backend 中完成，且不调用正式模型后端。本轮不触碰并行修改的 `lime-rs/src/app/runner.rs`、`lime-rs/src/commands/mod.rs`、live Provider smoke 写集，也不修改 `local_data_source.rs`。
@@ -1352,7 +1352,6 @@ pub fn run() {
            commands::aster_agent_cmd::action_runtime::agent_runtime_delete_session,
            commands::aster_agent_cmd::action_runtime::agent_runtime_respond_action,
            commands::aster_agent_cmd::tool_runtime::social_tools::social_generate_cover_image_cmd,
-            commands::theme_context_cmd::aster_agent_theme_context_search,
            // Models config commands
            commands::models_cmd::get_models_config,
            commands::models_cmd::save_models_config,
@@ -8083,7 +8083,6 @@ mod tests {
            "src/commands/aster_agent_cmd/action_runtime.rs",
            "src/commands/agent_cmd.rs",
            "src/commands/persona_cmd.rs",
-            "src/commands/theme_context_cmd.rs",
        ]
        .into_iter()
        .map(ToString::to_string)
@@ -53,7 +53,6 @@ pub mod skill_error;
 pub mod skill_exec_cmd;
 pub mod telegram_remote_cmd;
 pub mod telemetry_cmd;
-pub mod theme_context_cmd;
 pub mod tray_cmd;
 pub mod unified_memory_cmd;
 pub mod update_cmd;
@@ -1,530 +1,5 @@
-//! 主题工作台上下文搜索命令
+//! Retired legacy theme context search command.
 //!
-//! 为左侧栏“搜索上下文”提供真正的后端检索能力，
-//! 统一复用 Aster Agent + WebSearch 策略，并返回结构化结果。
-
-use crate::agent::{build_auxiliary_session_config, AsterAgentState, AsterAgentWrapper};
-use crate::config::GlobalConfigManagerState;
-use crate::database::DbConnection;
-use crate::services::memory_profile_prompt_service::{
-    merge_system_prompt_with_memory_context, MemoryPromptContext,
-};
-use crate::services::web_search_prompt_service::merge_system_prompt_with_web_search;
-use crate::services::web_search_runtime_service::apply_web_search_runtime_env;
-use crate::services::workspace_health_service::ensure_workspace_ready_with_auto_relocate;
-use crate::workspace::WorkspaceManager;
-use lime_agent::{
-    merge_system_prompt_with_runtime_agents, resolve_request_tool_policy_with_mode,
-    stream_reply_with_policy, RequestToolPolicyMode,
-};
-use serde::{Deserialize, Serialize};
-use std::path::Path;
-use tauri::State;
-use url::Url;
-use uuid::Uuid;
-
-const CONTEXT_SEARCH_SESSION_PREFIX: &str = "__lime_theme_context_search__";
-const FALLBACK_SUMMARY_LENGTH: usize = 420;
-
-#[derive(Debug, Clone, Copy, Deserialize, Serialize, PartialEq, Eq)]
-#[serde(rename_all = "lowercase")]
-pub enum ThemeContextSearchMode {
-    Web,
-    Social,
-}
-
-#[derive(Debug, Clone, Deserialize)]
-pub struct ThemeContextSearchRequest {
-    #[serde(alias = "workspaceId")]
-    pub workspace_id: String,
-    #[serde(default, alias = "projectId")]
-    pub project_id: Option<String>,
-    #[serde(alias = "providerType", alias = "providerId")]
-    pub provider_type: String,
-    pub model: String,
-    pub query: String,
-    pub mode: ThemeContextSearchMode,
-}
-
-#[derive(Debug, Clone, Serialize, Deserialize, PartialEq, Eq)]
-#[serde(rename_all = "camelCase")]
-pub struct ThemeContextSearchCitation {
-    pub title: String,
-    pub url: String,
-}
-
-#[derive(Debug, Clone, Serialize, Deserialize, PartialEq, Eq)]
-#[serde(rename_all = "camelCase")]
-pub struct ThemeContextSearchResponse {
-    pub title: String,
-    pub summary: String,
-    pub citations: Vec<ThemeContextSearchCitation>,
-    pub raw_response: String,
-    pub attempts_summary: String,
-}
-
-#[derive(Debug, Clone)]
-struct ParsedThemeContextSearchPayload {
-    title: Option<String>,
-    summary: Option<String>,
-    citations: Vec<ThemeContextSearchCitation>,
-}
-
-fn normalize_whitespace(value: &str) -> String {
-    value.split_whitespace().collect::<Vec<_>>().join(" ")
-}
-
-fn build_context_search_prompt(query: &str, mode: ThemeContextSearchMode) -> String {
-    let social_constraint = match mode {
-        ThemeContextSearchMode::Social => [
-            "优先寻找社交媒体平台、品牌官方账号、媒体社媒账号、KOL/KOC 讨论与趋势帖相关信息。",
-            "如果直接社媒来源不足，可补充官方网站或媒体报道，但摘要必须保留社媒传播视角。",
-            "适当优先关注小红书、微博、公众号、抖音、B站、知乎等中文平台。",
-        ]
-        .join("\n"),
-        ThemeContextSearchMode::Web => {
-            "优先提供最新且可信的公开网络资料，兼顾官方来源与主流媒体。".to_string()
-        }
-    };
-
-    [
-        "你是 Lime 的资料检索助手。",
-        "请先执行联网搜索，再输出整理结果。",
-        "你必须返回且仅返回一个 JSON 对象，不要使用 Markdown 代码块，不要输出多余说明。",
-        "JSON 结构如下：",
-        r#"{"title":"","summary":"","citations":[{"title":"","url":""}]}"#,
-        "字段要求：",
-        "1. title：12-28 字中文标题，概括本次检索主题。",
-        "2. summary：180-320 字中文摘要，聚合 3-5 个来源，突出时间点、关键事实、趋势或洞察。",
-        "3. citations：保留 3-5 条最重要来源，必须带可访问 URL。",
-        social_constraint.as_str(),
-        &format!("检索主题：{}", query.trim()),
-    ]
-    .join("\n")
-}
-
-fn strip_code_fence(value: &str) -> String {
-    value
-        .trim()
-        .trim_start_matches("```json")
-        .trim_start_matches("```")
-        .trim_end_matches("```")
-        .trim()
-        .to_string()
-}
-
-fn build_citation_title_from_url(url: &str) -> String {
-    Url::parse(url)
-        .ok()
-        .and_then(|parsed| {
-            parsed
-                .host_str()
-                .map(|host| host.trim_start_matches("www.").to_string())
-        })
-        .filter(|host| !host.is_empty())
-        .unwrap_or_else(|| "来源链接".to_string())
-}
-
-fn sanitize_url(url: &str) -> String {
-    url.trim_end_matches(&[',', ')', '.', ';', '!', '?'][..])
-        .trim()
-        .to_string()
-}
-
-fn parse_json_payload(raw_response: &str) -> Option<ParsedThemeContextSearchPayload> {
-    let trimmed = raw_response.trim();
-    if trimmed.is_empty() {
-        return None;
-    }
-
-    let fenced_match = regex::Regex::new(r"```(?:json)?\s*([\s\S]*?)\s*```")
-        .ok()
-        .and_then(|regex| regex.captures(trimmed))
-        .and_then(|captures| captures.get(1).map(|value| value.as_str().to_string()));
-    let json_block_match = regex::Regex::new(r"\{[\s\S]*\}")
-        .ok()
-        .and_then(|regex| regex.find(trimmed))
-        .map(|value| value.as_str().to_string());
-
-    let mut candidates = Vec::new();
-    if let Some(value) = fenced_match {
-        candidates.push(value);
-    }
-    if let Some(value) = json_block_match {
-        candidates.push(value);
-    }
-    candidates.push(trimmed.to_string());
-
-    for candidate in candidates {
-        let normalized = strip_code_fence(&candidate);
-        let parsed = match serde_json::from_str::<serde_json::Value>(&normalized) {
-            Ok(value) => value,
-            Err(_) => continue,
-        };
-
-        let citations_raw = parsed
-            .get("citations")
-            .and_then(serde_json::Value::as_array)
-            .or_else(|| parsed.get("sources").and_then(serde_json::Value::as_array));
-
-        let citations = citations_raw
-            .into_iter()
-            .flatten()
-            .filter_map(|item| {
-                let url = item
-                    .get("url")
-                    .and_then(serde_json::Value::as_str)
-                    .map(str::trim)
-                    .filter(|value| !value.is_empty())?;
-                let title = item
-                    .get("title")
-                    .and_then(serde_json::Value::as_str)
-                    .or_else(|| item.get("name").and_then(serde_json::Value::as_str))
-                    .map(normalize_whitespace)
-                    .filter(|value| !value.is_empty())
-                    .unwrap_or_else(|| build_citation_title_from_url(url));
-                Some(ThemeContextSearchCitation {
-                    title,
-                    url: url.to_string(),
-                })
-            })
-            .take(5)
-            .collect::<Vec<_>>();
-
-        let title = parsed
-            .get("title")
-            .and_then(serde_json::Value::as_str)
-            .map(normalize_whitespace)
-            .filter(|value| !value.is_empty());
-        let summary = parsed
-            .get("summary")
-            .and_then(serde_json::Value::as_str)
-            .or_else(|| parsed.get("content").and_then(serde_json::Value::as_str))
-            .map(normalize_whitespace)
-            .filter(|value| !value.is_empty());
-
-        return Some(ParsedThemeContextSearchPayload {
-            title,
-            summary,
-            citations,
-        });
-    }
-
-    None
-}
-
-fn extract_citations_from_text(raw_response: &str) -> Vec<ThemeContextSearchCitation> {
-    let mut citations = Vec::new();
-    let mut seen = std::collections::HashSet::new();
-
-    if let Ok(markdown_regex) = regex::Regex::new(r"\[([^\]]+)\]\((https?://[^\s)]+)\)") {
-        for captures in markdown_regex.captures_iter(raw_response) {
-            let url = captures
-                .get(2)
-                .map(|value| sanitize_url(value.as_str()))
-                .unwrap_or_default();
-            if url.is_empty() || !seen.insert(url.clone()) {
-                continue;
-            }
-            let title = captures
-                .get(1)
-                .map(|value| normalize_whitespace(value.as_str()))
-                .filter(|value| !value.is_empty())
-                .unwrap_or_else(|| build_citation_title_from_url(&url));
-            citations.push(ThemeContextSearchCitation { title, url });
-            if citations.len() >= 5 {
-                return citations;
-            }
-        }
-    }
-
-    if let Ok(url_regex) = regex::Regex::new(r"https?://[^\s)\]]+") {
-        for captures in url_regex.find_iter(raw_response) {
-            let url = sanitize_url(captures.as_str());
-            if url.is_empty() || !seen.insert(url.clone()) {
-                continue;
-            }
-            citations.push(ThemeContextSearchCitation {
-                title: build_citation_title_from_url(&url),
-                url,
-            });
-            if citations.len() >= 5 {
-                break;
-            }
-        }
-    }
-
-    citations
-}
-
-fn build_fallback_summary(raw_response: &str) -> String {
-    let without_citations = regex::Regex::new(r#""citations"\s*:\s*\[[\s\S]*?\]"#)
-        .ok()
-        .map(|regex| {
-            regex
-                .replace_all(&strip_code_fence(raw_response), "")
-                .to_string()
-        })
-        .unwrap_or_else(|| strip_code_fence(raw_response));
-    let normalized =
-        normalize_whitespace(&without_citations.replace(&['{', '}', '[', ']', '"'][..], " "));
-
-    if normalized.is_empty() {
-        return "暂无可用摘要，请重新尝试检索。".to_string();
-    }
-
-    if normalized.chars().count() <= FALLBACK_SUMMARY_LENGTH {
-        return normalized;
-    }
-
-    let mut summary = normalized
-        .chars()
-        .take(FALLBACK_SUMMARY_LENGTH)
-        .collect::<String>();
-    summary.push_str("...");
-    summary
-}
-
-fn build_fallback_title(query: &str, mode: ThemeContextSearchMode) -> String {
-    let suffix = match mode {
-        ThemeContextSearchMode::Social => "社媒搜索上下文",
-        ThemeContextSearchMode::Web => "网络搜索上下文",
-    };
-    format!("{} · {}", query.trim(), suffix)
-}
-
-fn normalize_search_result(
-    raw_response: &str,
-    query: &str,
-    mode: ThemeContextSearchMode,
-    attempts_summary: String,
-) -> ThemeContextSearchResponse {
-    let parsed = parse_json_payload(raw_response);
-    let citations = parsed
-        .as_ref()
-        .filter(|payload| !payload.citations.is_empty())
-        .map(|payload| payload.citations.clone())
-        .unwrap_or_else(|| extract_citations_from_text(raw_response));
-
-    ThemeContextSearchResponse {
-        title: parsed
-            .as_ref()
-            .and_then(|payload| payload.title.clone())
-            .unwrap_or_else(|| build_fallback_title(query, mode)),
-        summary: parsed
-            .as_ref()
-            .and_then(|payload| payload.summary.clone())
-            .unwrap_or_else(|| build_fallback_summary(raw_response)),
-        citations,
-        raw_response: raw_response.to_string(),
-        attempts_summary,
-    }
-}
-
-#[tauri::command]
-pub async fn aster_agent_theme_context_search(
-    state: State<'_, AsterAgentState>,
-    db: State<'_, DbConnection>,
-    config_manager: State<'_, GlobalConfigManagerState>,
-    request: ThemeContextSearchRequest,
-) -> Result<ThemeContextSearchResponse, String> {
-    let workspace_id = request.workspace_id.trim().to_string();
-    if workspace_id.is_empty() {
-        return Err("workspace_id 必填，请先选择项目工作区".to_string());
-    }
-
-    let provider_type = request.provider_type.trim().to_string();
-    if provider_type.is_empty() {
-        return Err("当前未选择可用模型，无法执行联网搜索".to_string());
-    }
-
-    let model = request.model.trim().to_string();
-    if model.is_empty() {
-        return Err("当前未选择可用模型，无法执行联网搜索".to_string());
-    }
-
-    let query = request.query.trim().to_string();
-    if query.is_empty() {
-        return Err("搜索词不能为空".to_string());
-    }
-
-    if !state.is_initialized().await {
-        state.init_agent_with_db(&db).await?;
-    }
-
-    let manager = WorkspaceManager::new(db.inner().clone());
-    let workspace = manager
-        .get(&workspace_id)
-        .map_err(|error| format!("读取 workspace 失败: {error}"))?
-        .ok_or_else(|| format!("Workspace 不存在: {workspace_id}"))?;
-    let ensured = ensure_workspace_ready_with_auto_relocate(&manager, &workspace)?;
-    let workspace_root = ensured.root_path.to_string_lossy().to_string();
-
-    let runtime_config = config_manager.config();
-    apply_web_search_runtime_env(&runtime_config);
-
-    let project_prompt = request
-        .project_id
-        .as_deref()
-        .map(str::trim)
-        .filter(|value| !value.is_empty())
-        .and_then(|project_id| {
-            match AsterAgentState::build_project_system_prompt(&db, project_id) {
-                Ok(prompt) => Some(prompt),
-                Err(error) => {
-                    tracing::warn!(
-                        "[ThemeContextSearch] 加载项目上下文失败，降级为基础搜索提示词: {}",
-                        error
-                    );
-                    None
-                }
-            }
-        });
-
-    let request_tool_policy = resolve_request_tool_policy_with_mode(
-        Some(true),
-        Some(RequestToolPolicyMode::Required),
-        false,
-    );
-    let system_prompt = lime_agent::merge_system_prompt_with_request_tool_policy(
-        merge_system_prompt_with_web_search(
-            merge_system_prompt_with_memory_context(
-                merge_system_prompt_with_runtime_agents(
-                    project_prompt,
-                    Some(Path::new(&workspace_root)),
-                ),
-                &runtime_config,
-                MemoryPromptContext::with_working_dir(Path::new(&workspace_root)),
-            ),
-            &runtime_config,
-        ),
-        &request_tool_policy,
-    );
-
-    let session_id = format!("{}-{}", CONTEXT_SEARCH_SESSION_PREFIX, Uuid::new_v4());
-    state
-        .configure_provider_from_pool(&db, &provider_type, &model, &session_id, None)
-        .await?;
-
-    let cancel_token = state.create_cancel_token(&session_id).await;
-    let execution_result = {
-        let agent_arc = state.get_agent_arc();
-        let guard = agent_arc.read().await;
-        let agent = guard
-            .as_ref()
-            .ok_or_else(|| "Agent not initialized".to_string())?;
-
-        stream_reply_with_policy(
-            agent,
-            &build_context_search_prompt(&query, request.mode),
-            Some(Path::new(&workspace_root)),
-            build_auxiliary_session_config(&session_id, system_prompt, false),
-            Some(cancel_token.clone()),
-            &request_tool_policy,
-            |_| {},
-        )
-        .await
-    };
-
-    state.remove_cancel_token(&session_id).await;
-    if let Err(error) = AsterAgentWrapper::delete_session(&db, &session_id).await {
-        tracing::warn!(
-            "[ThemeContextSearch] 删除临时会话失败: session={}, error={}",
-            session_id,
-            error
-        );
-    }
-
-    let execution = execution_result.map_err(|error| error.message)?;
-    let raw_response = execution.text_output.trim().to_string();
-    if raw_response.is_empty() {
-        return Err("上下文搜索未返回可用内容，请重试".to_string());
-    }
-
-    Ok(normalize_search_result(
-        &raw_response,
-        &query,
-        request.mode,
-        execution.attempts_summary,
-    ))
-}
-
-#[cfg(test)]
-mod tests {
-    use super::*;
-
-    #[test]
-    fn should_parse_json_result() {
-        let result = normalize_search_result(
-            r#"{"title":"智能体市场观察","summary":"市场讨论聚焦推理成本、工作流平台和企业落地节奏。","citations":[{"title":"官方博客","url":"https://example.com/blog"}]}"#,
-            "智能体市场 2026",
-            ThemeContextSearchMode::Web,
-            "WebSearch#1:success".to_string(),
-        );
-
-        assert_eq!(result.title, "智能体市场观察");
-        assert!(result.summary.contains("推理成本"));
-        assert_eq!(
-            result.citations,
-            vec![ThemeContextSearchCitation {
-                title: "官方博客".to_string(),
-                url: "https://example.com/blog".to_string(),
-            }]
-        );
-        assert_eq!(result.attempts_summary, "WebSearch#1:success");
-    }
-
-    #[test]
-    fn should_fallback_to_text_and_links_when_json_invalid() {
-        let result = normalize_search_result(
-            [
-                "2026 年社交媒体讨论聚焦 Agent 产品的真实 ROI。",
-                "参考链接：",
-                "[小红书热议](https://example.com/xhs)",
-                "https://example.com/weibo",
-            ]
-            .join("\n")
-            .as_str(),
-            "Agent 社媒讨论",
-            ThemeContextSearchMode::Social,
-            "WebSearch#1:success".to_string(),
-        );
-
-        assert!(result.title.contains("Agent 社媒讨论"));
-        assert!(result.summary.contains("真实 ROI"));
-        assert_eq!(
-            result.citations,
-            vec![
-                ThemeContextSearchCitation {
-                    title: "小红书热议".to_string(),
-                    url: "https://example.com/xhs".to_string(),
-                },
-                ThemeContextSearchCitation {
-                    title: "example.com".to_string(),
-                    url: "https://example.com/weibo".to_string(),
-                },
-            ]
-        );
-    }
-
-    #[test]
-    fn should_deserialize_theme_context_request_with_aliases() {
-        let request: ThemeContextSearchRequest = serde_json::from_str(
-            r#"{
-                "workspaceId": "workspace-test",
-                "projectId": "project-test",
-                "providerType": "openai",
-                "model": "gpt-4.1",
-                "query": "AI Agent 最新动态",
-                "mode": "web"
-            }"#,
-        )
-        .expect("request should deserialize");
-
-        assert_eq!(request.workspace_id, "workspace-test");
-        assert_eq!(request.project_id.as_deref(), Some("project-test"));
-        assert_eq!(request.provider_type, "openai");
-        assert_eq!(request.model, "gpt-4.1");
-        assert_eq!(request.mode, ThemeContextSearchMode::Web);
-    }
-}
+//! The active path is the App Server current session turn chain:
+//! `src/lib/api/themeContextSearch.ts -> agentSession/start -> agentSession/turn/start`.
+//! Do not re-register this module in the legacy desktop facade.
@@ -2322,9 +2322,9 @@ async function main() {
    summary.liveWebTurn = liveWebCompleted.turn;
    summary.liveWebTurnStatus = liveWebCompleted.turn?.status || null;
    summary.liveWebToolEvidence = liveWebCompleted.toolEvidence;
-    if (!summary.liveWebToolEvidence?.allRequiredCompletedForTurn) {
+    if (!summary.liveWebToolEvidence?.allRequiredOutputPresentForTurn) {
      const settledToolEvidence = await waitForCondition(
-        "等待 read model 出现 live WebSearch/WebFetch 工具事实",
+        "等待 read model 出现 live WebSearch/WebFetch 工具输出事实",
        async () => {
          const session = await readAppServerSession(
            options,
@@ -2338,7 +2338,7 @@ async function main() {
            passed: false,
            latest: toolEvidence,
          };
-          return toolEvidence.allRequiredCompletedForTurn
+          return toolEvidence.allRequiredOutputPresentForTurn
            ? { session, toolEvidence }
            : null;
        },
@@ -164,6 +164,10 @@ describe("live-provider-smoke-gate", () => {
    expect(content).toContain("liveWebToolEvidenceFromSession");
    expect(content).toContain("turnId: liveWebTurnId");
    expect(content).toContain("allRequiredCompletedForTurn");
+    expect(content).toContain("allRequiredOutputPresentForTurn");
+    expect(content).toContain(
+      "等待 read model 出现 live WebSearch/WebFetch 工具输出事实",
+    );
    expect(content).toContain("summary.assertions.liveWebSearchCompleted");
    expect(content).toContain("summary.assertions.liveWebFetchCompleted");
    expect(content).toContain(
@@ -4053,6 +4053,21 @@
      ],
      "allowedPaths": []
    },
+    {
+      "id": "rust-theme-context-search-legacy-command",
+      "classification": "dead",
+      "description": "旧主题上下文搜索 Tauri 命令已下线；current 事实源是 App Server agentSession current turn 网关",
+      "patterns": [
+        "pub mod theme_context_cmd;",
+        "commands::theme_context_cmd::",
+        "aster_agent_theme_context_search"
+      ],
+      "includePathPrefixes": [
+        "lime-rs/src",
+        "lime-rs/crates"
+      ],
+      "allowedPaths": []
+    },
    {
      "id": "rust-general-chat-dao",
      "classification": "dead-candidate",