补齐 Claude 上游错误展示与 MCP 并发隔离
Project CI / AI game creator shell Rust crates (push) Successful in 3m15s
Project CI / AI game creator shell Rust lane 2/2 (push) Successful in 4m30s
Project CI / Backend tests (push) Failing after 4m50s
Project CI / AI game creator shell Rust lane 1/2 (push) Successful in 6m23s
Project CI / Frontend tests (push) Successful in 3m38s
Project CI / Native shell tests (push) Has been cancelled
Project CI / Repository checks (push) Has been cancelled
Project CI / AI game creator shell web tests (push) Has been cancelled
Project CI / AI game creator shell Rust crates (push) Successful in 3m15s
Project CI / AI game creator shell Rust lane 2/2 (push) Successful in 4m30s
Project CI / Backend tests (push) Failing after 4m50s
Project CI / AI game creator shell Rust lane 1/2 (push) Successful in 6m23s
Project CI / Frontend tests (push) Successful in 3m38s
Project CI / Native shell tests (push) Has been cancelled
Project CI / Repository checks (push) Has been cancelled
Project CI / AI game creator shell web tests (push) Has been cancelled
前端保留 401/403/408/429/5xx 状态并补回归测试 按项目隔离 Claude MCP 注册表并防止旧回合误停新桥 更新 DirectProject 排障记录
This commit is contained in:
@@ -94,11 +94,14 @@ impl Drop for ActiveClaudeCodeTurnGuard {
|
||||
}
|
||||
}
|
||||
|
||||
struct ExternalMcpStopGuard;
|
||||
struct ExternalMcpStopGuard {
|
||||
root: PathBuf,
|
||||
token: String,
|
||||
}
|
||||
|
||||
impl Drop for ExternalMcpStopGuard {
|
||||
fn drop(&mut self) {
|
||||
stop_external_mcp_loopback();
|
||||
stop_external_mcp_loopback_for_root(&self.root, &self.token);
|
||||
}
|
||||
}
|
||||
|
||||
@@ -963,7 +966,10 @@ pub(crate) async fn direct_game_creator_claude_code_chat_at(
|
||||
let (mcp_url, mcp_token) =
|
||||
start_external_mcp_loopback_for_direct_turn(root, llm.web_search_enabled).await?;
|
||||
direct_turn_trace("claude-mcp-loopback-ready");
|
||||
let _mcp_guard = ExternalMcpStopGuard;
|
||||
let _mcp_guard = ExternalMcpStopGuard {
|
||||
root: root.to_path_buf(),
|
||||
token: mcp_token.clone(),
|
||||
};
|
||||
let home = crate::project::resolve_local_project_path(root, ".agent/runtime/claude-code/home")?;
|
||||
std::fs::create_dir_all(&home)
|
||||
.map_err(|error| format!("创建 cc Direct 隔离目录失败:{error}"))?;
|
||||
|
||||
@@ -35,12 +35,15 @@ const EXTERNAL_MCP_RESPONSE_SUMMARY_MAX_CHARS: usize = 240;
|
||||
const EXTERNAL_MCP_JOURNAL_MAX_BYTES: u64 = 8 * 1024 * 1024;
|
||||
const EXTERNAL_MCP_JOURNAL_RELATIVE_PATH: &str = ".agent/conversations/codex-responses.jsonl";
|
||||
static EXTERNAL_MCP_JOURNAL_LOCK: OnceLock<Mutex<()>> = OnceLock::new();
|
||||
static EXTERNAL_MCP_SERVER: OnceLock<Mutex<Option<ExternalMcpServer>>> = OnceLock::new();
|
||||
static EXTERNAL_MCP_SERVER: OnceLock<
|
||||
Mutex<std::collections::BTreeMap<PathBuf, ExternalMcpServer>>,
|
||||
> = OnceLock::new();
|
||||
tokio::task_local! {
|
||||
static EXTERNAL_MCP_BRIDGE_URL: String;
|
||||
}
|
||||
|
||||
pub(crate) struct ExternalMcpServer {
|
||||
token: String,
|
||||
_direct_turn_guard: Option<super::direct_tool_bridge::DirectToolBridgeTurnGuard>,
|
||||
_bridge: super::direct_tool_bridge::DirectToolBridge,
|
||||
task: tokio::task::JoinHandle<()>,
|
||||
@@ -1800,7 +1803,7 @@ async fn start_external_mcp_loopback_with_mode(
|
||||
};
|
||||
let state = ExternalMcpHttpState {
|
||||
bridge_url: bridge.url().to_string(),
|
||||
root,
|
||||
root: root.clone(),
|
||||
token: token.clone(),
|
||||
session_user_id: session.user_id,
|
||||
session_identity_generation: session.identity_generation,
|
||||
@@ -1814,25 +1817,40 @@ async fn start_external_mcp_loopback_with_mode(
|
||||
let _ = axum::serve(listener, app).await;
|
||||
});
|
||||
let url = format!("http://127.0.0.1:{}{route}", address.port());
|
||||
let registry = EXTERNAL_MCP_SERVER.get_or_init(|| Mutex::new(None));
|
||||
let registry =
|
||||
EXTERNAL_MCP_SERVER.get_or_init(|| Mutex::new(std::collections::BTreeMap::new()));
|
||||
let mut guard = registry
|
||||
.lock()
|
||||
.map_err(|_| "客户端 MCP 服务注册表不可用".to_string())?;
|
||||
if let Some(previous) = guard.take() {
|
||||
drop(previous);
|
||||
}
|
||||
*guard = Some(ExternalMcpServer {
|
||||
_direct_turn_guard: direct_turn_guard,
|
||||
_bridge: bridge,
|
||||
task,
|
||||
});
|
||||
guard.insert(
|
||||
root,
|
||||
ExternalMcpServer {
|
||||
token: token.clone(),
|
||||
_direct_turn_guard: direct_turn_guard,
|
||||
_bridge: bridge,
|
||||
task,
|
||||
},
|
||||
);
|
||||
Ok((url, token))
|
||||
}
|
||||
|
||||
pub(crate) fn stop_external_mcp_loopback() {
|
||||
if let Some(registry) = EXTERNAL_MCP_SERVER.get() {
|
||||
if let Ok(mut guard) = registry.lock() {
|
||||
guard.take();
|
||||
guard.clear();
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
pub(crate) fn stop_external_mcp_loopback_for_root(root: &Path, token: &str) {
|
||||
let Ok(root) = root.canonicalize() else {
|
||||
return;
|
||||
};
|
||||
if let Some(registry) = EXTERNAL_MCP_SERVER.get() {
|
||||
if let Ok(mut guard) = registry.lock() {
|
||||
if guard.get(&root).is_some_and(|server| server.token == token) {
|
||||
guard.remove(&root);
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
+18
-3
@@ -96,9 +96,7 @@ function directModelCallText(kind: ModelCallKind | null | undefined): string {
|
||||
case 'upstreamFailed':
|
||||
return (
|
||||
(kind.native ? directNativeKindText(kind.native) : null) ??
|
||||
(kind.statusCode >= 500
|
||||
? '智能服务暂时不可用,请稍后重试'
|
||||
: '智能创作请求无效,请稍后重试')
|
||||
directUpstreamStatusText(kind.statusCode)
|
||||
);
|
||||
case 'paidCreditsInsufficient':
|
||||
return '泥点余额不足,本轮游戏生成已中断。请充值后发送“继续”,系统会从当前项目进度接着完成。';
|
||||
@@ -113,6 +111,23 @@ function directModelCallText(kind: ModelCallKind | null | undefined): string {
|
||||
}
|
||||
}
|
||||
|
||||
function directUpstreamStatusText(statusCode: number): string {
|
||||
switch (statusCode) {
|
||||
case 401:
|
||||
return '上游服务拒绝认证(HTTP 401),请重新登录后重试';
|
||||
case 403:
|
||||
return '上游服务拒绝访问(HTTP 403),请检查账号权限后重试';
|
||||
case 408:
|
||||
return '上游请求超时(HTTP 408),请稍后重试';
|
||||
case 429:
|
||||
return '上游服务繁忙(HTTP 429),请稍后重试';
|
||||
default:
|
||||
return statusCode >= 500
|
||||
? `智能服务暂时不可用(HTTP ${statusCode}),请稍后重试`
|
||||
: `智能创作请求被上游拒绝(HTTP ${statusCode}),请检查请求后重试`;
|
||||
}
|
||||
}
|
||||
|
||||
function directTurnDeadlineText(deadline: Deadline): string {
|
||||
switch (deadline) {
|
||||
case 'response-idle':
|
||||
|
||||
@@ -0,0 +1,34 @@
|
||||
import { describe, expect, it } from 'vitest';
|
||||
|
||||
import { directTurnFailureNoticeText } from '../src/view/project-development/chat/conversation/directTurnFailure';
|
||||
|
||||
describe('DirectProject 上游失败文案', () => {
|
||||
it.each([401, 403, 408, 429, 500, 502, 503, 504])(
|
||||
'保留 HTTP %s 的具体原因',
|
||||
(statusCode) => {
|
||||
const text = directTurnFailureNoticeText({
|
||||
type: 'modelCallFailed',
|
||||
kind: { type: 'upstreamFailed', statusCode, native: null },
|
||||
detail: `上游 HTTP ${statusCode}`,
|
||||
});
|
||||
|
||||
expect(text).toContain(`HTTP ${statusCode}`);
|
||||
expect(text).not.toContain('智能创作请求无效');
|
||||
},
|
||||
);
|
||||
|
||||
it('原生分类优先于 HTTP 状态文案', () => {
|
||||
const text = directTurnFailureNoticeText({
|
||||
type: 'modelCallFailed',
|
||||
kind: {
|
||||
type: 'upstreamFailed',
|
||||
statusCode: 429,
|
||||
native: { type: 'context-window-exceeded' },
|
||||
},
|
||||
detail: 'upstream',
|
||||
});
|
||||
|
||||
expect(text).toContain('模型上下文已超限');
|
||||
expect(text).not.toContain('HTTP 429');
|
||||
});
|
||||
});
|
||||
@@ -6290,7 +6290,8 @@ Cocos Creator 根目录由 `package.json.creator.version` 与普通 `assets/`
|
||||
- **根因 5(验收反馈复用 assistant ID)**:同一 client turn 进入 `ReviewRequired` 后会再次调用 cc。首次回复已经占用 `direct-codex:<clientTurnId>:assistant`,第二次不同正文沿用该 ID 会被历史层正确拒绝为冲突,随后却被错误投影成 `runtime-unclassified`。真实诊断中可见「写入本项目对话历史失败:…assistant」且历史已经有该条回复。
|
||||
- **根因 6(cc 字符串错误覆盖了上游分类)**:DirectProject 的 Claude Code 路由原本把 sidecar 返回的所有字符串都包装成 `LlmError::Transport`,因此 HTTP 429、401、408、5xx、sidecar 超时、空回执和无效 JSON 都显示成「执行通道未能建立或已断开」。
|
||||
- **根因 7(内部 Claude MCP 桥误用外部只读模式)**:Claude Code sidecar 使用的 loopback MCP 原本调用 `start_external_client_tool_bridge(..., false)`,桥状态 `direct_turn_execution=false` 且没有 `begin_user_turn()` 授权;`agc_register_delivery_contract`、`agc_delivery_status`、`agc_update_plan` 每次都会返回 `ToolRequiresDirectTurn`,模型收到错误后又重复注册计划,最终陷入反馈死循环直到超时。
|
||||
- **现行口径**:cc 成功出口由放行侧补写 `DirectTurnTerminal::completed()`(`finish_if_unfinished` 幂等,codex 已写过终态时是空操作);cc 每次解析成功后都把实际落盘的回复 item id 同步下发 `ThreadEvent::item_completed(ThreadItem::Message{role:"assistant"})`。首个回复沿用 `direct-codex:<clientTurnId>:assistant`,同一回合的反馈回复遇到内容冲突时追加 `:assistant:<uuid>`,相同内容仍按原 ID 幂等;落盘失败按回合失败收口。Claude Code 的失败文本先投影到与 Codex 相同的 `LlmError` 分类:HTTP 状态、sidecar 超时、空回执和无效 JSON 分别复用上游、超时、空响应和反序列化语义;上游状态摘要与重试建议按状态码给出。内部 Claude Direct MCP 必须使用 `direct_turn_execution=true` 的工具桥并持有 `begin_user_turn()` guard;用户手动启动的外部 MCP 仍保持非 Direct 模式。sidecar 按 `mcp__<server>` 前缀整体放行请求里声明的 MCP 服务器(权限策略在宿主侧执行)。
|
||||
- **根因 8(MCP 全局单槽位)**:外部 MCP 注册表原本只有一个 `Option<ExternalMcpServer>`,不同项目的 Claude 回合会互相 abort;一个回合结束时的全局 stop 还可能误停另一个项目的桥。
|
||||
- **现行口径**:cc 成功出口由放行侧补写 `DirectTurnTerminal::completed()`(`finish_if_unfinished` 幂等,codex 已写过终态时是空操作);cc 每次解析成功后都把实际落盘的回复 item id 同步下发 `ThreadEvent::item_completed(ThreadItem::Message{role:"assistant"})`。首个回复沿用 `direct-codex:<clientTurnId>:assistant`,同一回合的反馈回复遇到内容冲突时追加 `:assistant:<uuid>`,相同内容仍按原 ID 幂等;落盘失败按回合失败收口。Claude Code 的失败文本先投影到与 Codex 相同的 `LlmError` 分类:HTTP 状态、sidecar 超时、空回执和无效 JSON 分别复用上游、超时、空响应和反序列化语义;上游状态摘要与重试建议按状态码给出。内部 Claude Direct MCP 必须使用 `direct_turn_execution=true` 的工具桥并持有 `begin_user_turn()` guard;用户手动启动的外部 MCP 仍保持非 Direct 模式。MCP 注册表按 canonical project root 分桶,停止操作再核对 server token,迟到的旧回合不能误停同项目的新桥。sidecar 按 `mcp__<server>` 前缀整体放行请求里声明的 MCP 服务器(权限策略在宿主侧执行)。
|
||||
- **诊断口径**:`agent.direct_turn.host_dropped` / `agent.direct_turn.panic` 里的令牌字段必须写 `tt=`,写 `turnToken=` 会命中脱敏标记,整行变成 `<sensitive diagnostic details redacted>`,离线只剩"说不出原因"的 HostDropped。
|
||||
- **验证**:dev 栈里用 CDP 注入真实回合(`node %TEMP%\agc-cdp.mjs <expr>`):①读文件轮 `claude-parse-done chars=108`,`.agent/conversations/project.jsonl` 出现 `direct-codex:cdp-…:assistant` 条目,回复内容与 `game/index.html` 前两行(`<!doctype html>` / `<html lang="zh-CN">`)逐字一致(证明宿主工具真的执行了);②聊天视图打开时注入 `只回三个字:收到了`,DOM 断言(`document.body.innerText`)同时出现用户气泡 `11:40:05`、助手回复 `收到了` 与 `本轮结束于 11:40:16 · 耗时 10.7秒`(证明 `item.completed` 实时投影生效,不必重进项目);同一日志不再出现新的 `host_dropped`。另有 `agent::claude_code_cli::tests::direct_claude_feedback_reply_does_not_fail_on_a_reused_client_turn_id` 回归覆盖同一回合两次不同回复。`cargo test … -- claude_code_cli::tests direct_turn_failure::tests` 19 passed。
|
||||
- **关联**:`apps/ai-game-creator-shell/src-tauri/src/agent/thread_manager/dispatch.rs`、`apps/ai-game-creator-shell/src-tauri/src/agent/direct_turn_failure.rs`、`apps/ai-game-creator-shell/src-tauri/src/agent/claude_code_cli.rs`、`apps/ai-game-creator-shell/agent-sidecar/src/index.mjs`。
|
||||
|
||||
Reference in New Issue
Block a user