feat: 命令台子代理链路(主任务/模型子代理/委派子任务 + 过程调用实时同步)

host:链路条目带 thread 归属;delegate 向父链路文件写 agent-start/done 标记。服务端:GET /tasks/{id}/agents 聚合父链路 + 子链路文件成代理链(状态口径:委派以结束标记为准、无标记即进行中;模型子代理无结束信号时父任务运行中为进行中、否则未知)。命令台:新增子代理链路区,随 SSE 流防抖刷新;扁平链路也渲染 agent 标记。
This commit is contained in:
host committed 2026-10-27 11:51:11 +08:00
1 parent 34cd0eeba7
commit c4d3ec9bff
6 files changed
+366 -14

No files matched your search

+48 -1
View File
@@ -248,6 +248,8 @@
<div class="bar"><button id="btnContinueConfirm" class="primary">继续会话</button><span id="continueMsgSpan" class="muted"></span></div> <div class="bar"><button id="btnContinueConfirm" class="primary">继续会话</button><span id="continueMsgSpan" class="muted"></span></div>
</div> </div>
<pre id="detailBody" style="display:none"></pre> <pre id="detailBody" style="display:none"></pre>
<h3>子代理链路</h3>
<div id="agentBox" class="muted">点任务行查看详情后,这里显示子代理链路</div>
<h3>实时链路</h3> <h3>实时链路</h3>
<div id="traceBox" class="muted">点任务行查看详情后,这里显示过程链路</div> <div id="traceBox" class="muted">点任务行查看详情后,这里显示过程链路</div>
<div id="costBox"></div> <div id="costBox"></div>
@@ -269,7 +271,7 @@ var $ = function (s) { return document.querySelector(s); };
var KEY_STORE = "codex-console-key"; var KEY_STORE = "codex-console-key";
var currentTask = null; var currentTask = null;
var traceAbort = null; var traceAbort = null;
var TRACE_KIND = { exec: "命令", file: "文件", call: "调用", tool: "工具", message: "消息", reasoning: "思考", plan: "计划" }; var TRACE_KIND = { exec: "命令", file: "文件", call: "调用", tool: "工具", message: "消息", reasoning: "思考", plan: "计划", "agent-start": "子代理", "agent-done": "子代理" };
var TRACE_LIMIT = 1000; var TRACE_LIMIT = 1000;
var TRACE_TERMINAL = ["completed", "failed", "interrupted", "merged", "merge-failed"]; var TRACE_TERMINAL = ["completed", "failed", "interrupted", "merged", "merge-failed"];
@@ -485,6 +487,7 @@ async function loadTasks() {
function stopTrace() { function stopTrace() {
if (traceAbort) { try { traceAbort.abort(); } catch (e) {} traceAbort = null; } if (traceAbort) { try { traceAbort.abort(); } catch (e) {} traceAbort = null; }
if (agentTimer) { clearTimeout(agentTimer); agentTimer = null; }
} }
function traceLine(e) { function traceLine(e) {
var k = TRACE_KIND[e.kind] || e.kind || "?"; var k = TRACE_KIND[e.kind] || e.kind || "?";
@@ -496,6 +499,8 @@ function traceLine(e) {
else if (e.kind === "tool") body = esc(e.server || "") + "/" + esc(e.tool || "") + " " + esc(e.status || ""); else if (e.kind === "tool") body = esc(e.server || "") + "/" + esc(e.tool || "") + " " + esc(e.status || "");
else if (e.kind === "call") body = esc(e.name || ""); else if (e.kind === "call") body = esc(e.name || "");
else if (e.kind === "plan") body = esc(e.text || ""); else if (e.kind === "plan") body = esc(e.text || "");
else if (e.kind === "agent-start") body = "开始:" + esc(e.name || "") + (e.prompt ? "(" + esc(e.prompt) + ")" : "");
else if (e.kind === "agent-done") body = "结束:" + esc(e.name || "") + " " + esc(e.status || "");
else body = esc(JSON.stringify(e).slice(0, 300)); else body = esc(JSON.stringify(e).slice(0, 300));
return "[" + k + "] " + body; return "[" + k + "] " + body;
} }
@@ -512,6 +517,44 @@ function renderTrace(entries, note, doneStatus) {
if (doneStatus) html += "<br><span class='muted'>链路已结束(" + esc(doneStatus) + ")</span>"; if (doneStatus) html += "<br><span class='muted'>链路已结束(" + esc(doneStatus) + ")</span>";
box.innerHTML = html; box.innerHTML = html;
} }
var AGENT_STATUS = { running: "进行中", completed: "已完成", failed: "失败", interrupted: "已中断", unknown: "未知" };
var AGENT_KIND = { main: "主任务", delegate: "委派子任务", model: "模型子代理" };
var AGENT_CALL_SHOWN = 50;
var agentTimer = null;
function renderAgents(data) {
var box = $("#agentBox");
var html = "";
if (!data) { box.innerHTML = "(暂无子代理链路)"; return; }
if (data.note) html += "<span class='err'>" + esc(data.note) + "</span><br>";
var agents = data.agents || [];
if (!agents.length) html += "(暂无子代理)";
agents.forEach(function (a) {
var st = AGENT_STATUS[a.status] || a.status || "?";
html += "<div style='margin:6px 0;padding:8px 10px;border:1px solid var(--border);border-radius:var(--radius);background:#fff'>";
html += "<b>" + esc(AGENT_KIND[a.kind] || a.kind || "") + ":</b>" + esc(a.name || "") +
" <span class='st-" + esc(a.status) + "'>" + esc(st) + "</span>" +
" | 过程调用 " + (a.calls ? a.calls.length : 0) + " 条" +
(a.updated_ts ? " | 最近活动 " + fmt(a.updated_ts / 1000) : "") + "<br>";
if (a.prompt) html += "<span class='muted'>任务:" + esc(a.prompt) + "</span><br>";
if (a.calls && a.calls.length) {
var shown = a.calls.slice(-AGENT_CALL_SHOWN);
if (a.calls.length > AGENT_CALL_SHOWN) html += "(仅显示最近 " + AGENT_CALL_SHOWN + " 条,共 " + a.calls.length + " 条)<br>";
html += shown.map(traceLine).join("<br>");
} else html += "<span class='muted'>(暂无过程调用)</span>";
html += "</div>";
});
box.innerHTML = html;
}
async function loadAgents(taskId) {
try {
var d = await api("/tasks/" + encodeURIComponent(taskId) + "/agents", "GET");
renderAgents(d);
} catch (e) { $("#agentBox").innerHTML = "<span class='err'>子代理链路加载失败:" + esc(e.message) + "</span>"; }
}
function scheduleAgents(taskId) {
if (agentTimer) clearTimeout(agentTimer);
agentTimer = setTimeout(function () { loadAgents(taskId); }, 1000);
}
async function startTrace(taskId, status) { async function startTrace(taskId, status) {
stopTrace(); stopTrace();
var box = $("#traceBox"); var box = $("#traceBox");
@@ -561,10 +604,12 @@ async function startTrace(taskId, status) {
entries.forEach(function (e, i) { index[e.id] = i; }); entries.forEach(function (e, i) { index[e.id] = i; });
note = m.note || ""; note = m.note || "";
renderTrace(entries, note, ""); renderTrace(entries, note, "");
scheduleAgents(taskId);
} else if (m.type === "entry") { } else if (m.type === "entry") {
var e = m.entry; var e = m.entry;
if (e.id in index) entries[index[e.id]] = e; else { index[e.id] = entries.length; entries.push(e); } if (e.id in index) entries[index[e.id]] = e; else { index[e.id] = entries.length; entries.push(e); }
renderTrace(entries, note, ""); renderTrace(entries, note, "");
scheduleAgents(taskId);
} else if (m.type === "done") { } else if (m.type === "done") {
done = true; doneStatus = m.status; done = true; doneStatus = m.status;
renderTrace(entries, note, doneStatus); renderTrace(entries, note, doneStatus);
@@ -618,6 +663,8 @@ async function showDetail(taskId) {
$("#continueMsgSpan").textContent = ""; $("#continueMsgSpan").textContent = "";
$("#continueForm").style.display = (TRACE_TERMINAL.indexOf(t.status) >= 0 && t.status !== "merged") ? "block" : "none"; $("#continueForm").style.display = (TRACE_TERMINAL.indexOf(t.status) >= 0 && t.status !== "merged") ? "block" : "none";
startTrace(taskId, t.status); startTrace(taskId, t.status);
$("#agentBox").innerHTML = "子代理链路加载中…";
loadAgents(taskId);
loadFollowups(taskId); loadFollowups(taskId);
} catch (e) { $("#detail").innerHTML = "<span class='err'>加载失败:" + esc(e.message) + "</span>"; } } catch (e) { $("#detail").innerHTML = "<span class='err'>加载失败:" + esc(e.message) + "</span>"; }
} }
+144
View File
@@ -184,6 +184,127 @@ def _read_trace_entries(path):
except OSError: except OSError:
return [] return []
return entries return entries
AGENT_CALL_LIMIT = 1000
def _read_agent_entries(path):
"""读链路文件供子代理聚合:按 (thread, id) 去重,后到的完成态
覆盖先到的开始态,保首次顺序;坏行与无 id 行跳过;文件不
存在为空。同一 item id 在不同线程出现时不互相覆盖(模型内
子代理与主任务各有自己的线程)。"""
entries = []
index = {}
try:
with open(path, "r", encoding="utf-8") as f:
for line in f:
line = line.strip()
if not line:
continue
try:
entry = json.loads(line)
except ValueError:
continue
if not isinstance(entry, dict) or "id" not in entry:
continue
key = (entry.get("thread") or "", entry["id"])
if key in index:
entries[index[key]] = entry
else:
index[key] = len(entries)
entries.append(entry)
except OSError:
return []
return entries
def _latest_ts(values):
vals = [v for v in values if isinstance(v, (int, float))]
return max(vals) if vals else None
def _agent_status_from_task(status):
return {"queued": "running", "running": "running",
"completed": "completed", "merged": "completed",
"failed": "failed", "merge-failed": "failed",
"interrupted": "interrupted"}.get(status, "unknown")
def _build_agent_chain(task, trace_dir):
"""任务 → 子代理链:主任务 + 模型内子代理(同文件多线程)+
委派子任务(父链路里的 agent-start/done 标记 + 子链路文件)。
状态诚实口径:委派子任务以 agent-done 标记为准,无标记即
进行中——委派进程与父任务相互独立,父任务结束不代表子任务
停;模型内子代理无明确结束信号,父任务运行中即进行中,
否则未知。无 thread 的旧条目一律归主任务。
"""
parent_path = os.path.join(
trace_dir, _trace_file_name(task["id"]) + ".jsonl")
entries = _read_agent_entries(parent_path)
calls = [e for e in entries
if e.get("kind") not in ("agent-start", "agent-done")]
main_thread = None
for e in calls:
if e.get("thread"):
main_thread = e["thread"]
break
root_calls = [e for e in calls
if not e.get("thread") or e.get("thread") == main_thread]
agents = [{
"id": "main", "name": "主任务", "kind": "main", "parent": None,
"status": _agent_status_from_task(task.get("status")),
"prompt": None, "trace": None,
"calls": root_calls[:AGENT_CALL_LIMIT],
"updated_ts": _latest_ts([e.get("ts") for e in root_calls]),
}]
task_running = task.get("status") in ("queued", "running")
threads = {}
for e in calls:
t = e.get("thread")
if t and t != main_thread:
threads.setdefault(t, []).append(e)
for t, es in threads.items():
agents.append({
"id": t, "name": "模型子代理 " + t[:12], "kind": "model",
"parent": "main",
"status": "running" if task_running else "unknown",
"prompt": None, "trace": None,
"calls": es[:AGENT_CALL_LIMIT],
"updated_ts": _latest_ts([e.get("ts") for e in es]),
})
starts = {}
dones = {}
for e in entries:
if e.get("kind") == "agent-start" and e.get("name"):
starts[e["name"]] = e
elif e.get("kind") == "agent-done" and e.get("name"):
dones[e["name"]] = e
for name, s in starts.items():
trace_name = s.get("trace") or dones.get(name, {}).get("trace")
child_calls = []
if trace_name:
child_path = os.path.join(
trace_dir, os.path.basename(trace_name))
child_calls = [e for e in _read_agent_entries(child_path)
if e.get("kind") not in
("agent-start", "agent-done")]
done = dones.get(name)
status = "running"
if done and done.get("status") in (
"completed", "failed", "interrupted"):
status = done["status"]
agents.append({
"id": "delegate:" + name, "name": name, "kind": "delegate",
"parent": "main", "status": status,
"prompt": s.get("prompt"), "trace": trace_name,
"calls": child_calls[:AGENT_CALL_LIMIT],
"updated_ts": _latest_ts(
[e.get("ts") for e in child_calls] +
[s.get("ts"), done.get("ts") if done else None]),
})
return agents
DEFAULT_SWEEP_THRESHOLD_SECONDS = 300 DEFAULT_SWEEP_THRESHOLD_SECONDS = 300
# 限流默认值:每 IP / 每 API key 每分钟 120 次;环境变量 # 限流默认值:每 IP / 每 API key 每分钟 120 次;环境变量
# CLOUD_RATE_LIMIT_PER_MINUTE / CLOUD_RATE_LIMIT_WINDOW_SECONDS 可调。 # CLOUD_RATE_LIMIT_PER_MINUTE / CLOUD_RATE_LIMIT_WINDOW_SECONDS 可调。
@@ -1262,6 +1383,8 @@ class CloudControlPlaneHandler(BaseHTTPRequestHandler):
user_id, rest[: -len("/trace/stream")]) user_id, rest[: -len("/trace/stream")])
elif rest.endswith("/trace") and method == "GET": elif rest.endswith("/trace") and method == "GET":
self._task_trace(user_id, rest[: -len("/trace")]) self._task_trace(user_id, rest[: -len("/trace")])
elif rest.endswith("/agents") and method == "GET":
self._task_agents(user_id, rest[: -len("/agents")])
elif rest.endswith("/cancel") and method == "POST": elif rest.endswith("/cancel") and method == "POST":
self._cancel_task(user_id, rest[: -len("/cancel")]) self._cancel_task(user_id, rest[: -len("/cancel")])
elif rest.endswith("/merge") and method == "POST": elif rest.endswith("/merge") and method == "POST":
@@ -2471,6 +2594,27 @@ class CloudControlPlaneHandler(BaseHTTPRequestHandler):
self._send_json(200, {"task_id": task_id, "status": task["status"], self._send_json(200, {"task_id": task_id, "status": task["status"],
"entries": entries, "note": note}) "entries": entries, "note": note})
def _task_agents(self, user_id, task_id):
row = self._fetch_task(task_id)
if row is None:
self._send_json(404, {"error": "task not found"})
return
task = _task_to_dict(row)
with _auth_store(self.server.db_path) as store:
if not store.check_permission(user_id, task["project_id"], "read"):
self._send_json(403, {"error": "forbidden"})
return
worktree_host = task.get("worktree_host")
if worktree_host and worktree_host != socket.gethostname():
self._send_json(200, {"task_id": task_id, "status": task["status"],
"agents": [],
"note": ("链路文件在 worktree 主机 %s 上,本机读不到"
% worktree_host)})
return
agents = _build_agent_chain(task, _trace_dir_for(self.server))
self._send_json(200, {"task_id": task_id, "status": task["status"],
"agents": agents, "note": ""})
def _task_trace_stream(self, user_id, task_id): def _task_trace_stream(self, user_id, task_id):
row = self._fetch_task(task_id) row = self._fetch_task(task_id)
if row is None: if row is None:
+1 -1
View File
@@ -64,7 +64,7 @@ class ConsoleTestCase(unittest.TestCase):
self.assertIn("Codex 命令台", text) self.assertIn("Codex 命令台", text)
# 壳必须引用它实际调的接口,防 JS 与服务端路由漂移。 # 壳必须引用它实际调的接口,防 JS 与服务端路由漂移。
for marker in ("/health", "/projects", "/tasks", "/workers", for marker in ("/health", "/projects", "/tasks", "/workers",
"X-API-Key"): "X-API-Key", "agentBox", "/agents"):
self.assertIn(marker, text) self.assertIn(marker, text)
def test_console_trailing_slash_same_page(self): def test_console_trailing_slash_same_page(self):
+73
View File
@@ -173,6 +173,79 @@ class TaskTraceTestCase(unittest.TestCase):
self.assertEqual(status, 401) self.assertEqual(status, 401)
self.assertEqual(events, []) self.assertEqual(events, [])
def test_agents_auth_and_not_found(self):
status, _ = self._req(
"GET", "/tasks/%s/agents" % self.task_id)
self.assertEqual(status, 401)
status, _ = self._req(
"GET", "/tasks/%s/agents" % self.task_id, key=self.other_key)
self.assertEqual(status, 403)
status, _ = self._req(
"GET", "/tasks/nope/agents", key=self.owner_key)
self.assertEqual(status, 404)
def test_agents_chain_groups_calls_by_agent(self):
self._set_task(status="running")
self._write_trace([
json.dumps({"kind": "exec", "id": "c1", "thread": "th-main",
"command": "make", "ts": 1000}),
json.dumps({"kind": "tool", "id": "t1", "thread": "th-sub",
"server": "s", "tool": "read", "ts": 1001}),
json.dumps({"kind": "agent-start", "id": "agent:rev",
"name": "rev", "prompt": "review code",
"trace": "rev.jsonl", "ts": 1002}),
json.dumps({"kind": "agent-done", "id": "agent:rev:done",
"name": "rev", "status": "completed",
"trace": "rev.jsonl", "ts": 1004}),
])
with open(os.path.join(self.trace_dir, "rev.jsonl"),
"w", encoding="utf-8") as f:
f.write(json.dumps({"kind": "exec", "id": "c9",
"command": "grep", "ts": 1003}) + "\n")
status, data = self._req(
"GET", "/tasks/%s/agents" % self.task_id, key=self.owner_key)
self.assertEqual(status, 200)
self.assertEqual(data["status"], "running")
by_kind = {a["kind"]: a for a in data["agents"]}
main = by_kind["main"]
self.assertEqual(main["status"], "running")
self.assertEqual([e["id"] for e in main["calls"]], ["c1"])
model = by_kind["model"]
self.assertEqual(model["status"], "running")
self.assertEqual([e["id"] for e in model["calls"]], ["t1"])
delegate = [a for a in data["agents"]
if a["kind"] == "delegate"][0]
self.assertEqual(delegate["name"], "rev")
self.assertEqual(delegate["status"], "completed")
self.assertEqual(delegate["prompt"], "review code")
self.assertEqual([e["id"] for e in delegate["calls"]], ["c9"])
self.assertEqual(delegate["updated_ts"], 1004)
def test_agents_delegate_without_done_stays_running(self):
# 委派进程与父任务相互独立:父任务已结束、无结束标记时,
# 子任务仍算进行中(不能谎报它已中断)。
self._set_task(status="completed")
self._write_trace([
json.dumps({"kind": "agent-start", "id": "agent:rev",
"name": "rev", "prompt": "review",
"trace": "rev.jsonl", "ts": 1002}),
])
status, data = self._req(
"GET", "/tasks/%s/agents" % self.task_id, key=self.owner_key)
self.assertEqual(status, 200)
delegate = [a for a in data["agents"]
if a["kind"] == "delegate"][0]
self.assertEqual(delegate["status"], "running")
self.assertEqual(data["agents"][0]["status"], "completed")
def test_agents_missing_file_is_main_only(self):
status, data = self._req(
"GET", "/tasks/%s/agents" % self.task_id, key=self.owner_key)
self.assertEqual(status, 200)
self.assertEqual(len(data["agents"]), 1)
self.assertEqual(data["agents"][0]["kind"], "main")
self.assertEqual(data["agents"][0]["calls"], [])
if __name__ == "__main__": if __name__ == "__main__":
unittest.main() unittest.main()
+57 -12
View File
@@ -214,6 +214,8 @@ def _summarize_trace_item(method, params):
返回 [] 表示该通知不进链路(userMessage 等)。 返回 [] 表示该通知不进链路(userMessage 等)。
ts 取通知的 completedAtMs/startedAtMs(毫秒 epoch), ts 取通知的 completedAtMs/startedAtMs(毫秒 epoch),
没有时为 None,服务端展示时按无时间处理。 没有时为 None,服务端展示时按无时间处理。
thread 取通知的 threadId(模型内子代理与主任务同文件时,
服务端凭它把条目归属到不同代理;没有时为 None,归主任务)。
""" """
item = (params or {}).get("item") item = (params or {}).get("item")
if not isinstance(item, dict): if not isinstance(item, dict):
@@ -221,8 +223,11 @@ def _summarize_trace_item(method, params):
item_type = item.get("type") item_type = item.get("type")
item_id = item.get("id") or "" item_id = item.get("id") or ""
ts = params.get("completedAtMs") or params.get("startedAtMs") ts = params.get("completedAtMs") or params.get("startedAtMs")
thread = params.get("threadId")
if not thread and isinstance(params.get("thread"), dict):
thread = params["thread"].get("id")
if item_type == "commandExecution": if item_type == "commandExecution":
return [{"kind": "exec", "id": item_id, "ts": ts, return [{"kind": "exec", "id": item_id, "ts": ts, "thread": thread,
"command": (item.get("command") or "")[:500], "command": (item.get("command") or "")[:500],
"cwd": item.get("cwd") or "", "cwd": item.get("cwd") or "",
"exitCode": item.get("exitCode"), "exitCode": item.get("exitCode"),
@@ -235,6 +240,7 @@ def _summarize_trace_item(method, params):
path = ch.get("path") or "" path = ch.get("path") or ""
entries.append({"kind": "file", entries.append({"kind": "file",
"id": item_id + ":" + path, "ts": ts, "id": item_id + ":" + path, "ts": ts,
"thread": thread,
"path": path, "change": ch.get("kind") or "", "path": path, "change": ch.get("kind") or "",
"status": item.get("status") or ""}) "status": item.get("status") or ""})
return entries return entries
@@ -242,32 +248,60 @@ def _summarize_trace_item(method, params):
text = item.get("text") or "" text = item.get("text") or ""
if not text: if not text:
return [] return []
return [{"kind": "message", "id": item_id, "ts": ts, return [{"kind": "message", "id": item_id, "ts": ts, "thread": thread,
"phase": item.get("phase") or "", "text": text[:300]}] "phase": item.get("phase") or "", "text": text[:300]}]
if item_type == "reasoning": if item_type == "reasoning":
summary = item.get("summary") or item.get("content") or [] summary = item.get("summary") or item.get("content") or []
text = " ".join(s for s in summary if isinstance(s, str)).strip() text = " ".join(s for s in summary if isinstance(s, str)).strip()
if not text: if not text:
return [] return []
return [{"kind": "reasoning", "id": item_id, "ts": ts, return [{"kind": "reasoning", "id": item_id, "ts": ts, "thread": thread,
"summary": text[:200]}] "summary": text[:200]}]
if item_type == "mcpToolCall": if item_type == "mcpToolCall":
return [{"kind": "tool", "id": item_id, "ts": ts, return [{"kind": "tool", "id": item_id, "ts": ts, "thread": thread,
"server": item.get("server") or "", "server": item.get("server") or "",
"tool": item.get("tool") or "", "tool": item.get("tool") or "",
"status": item.get("status") or ""}] "status": item.get("status") or ""}]
if item_type == "functionCallOutput": if item_type == "functionCallOutput":
return [{"kind": "call", "id": item_id, "ts": ts, return [{"kind": "call", "id": item_id, "ts": ts, "thread": thread,
"name": item.get("name") or ""}] "name": item.get("name") or ""}]
if item_type == "plan": if item_type == "plan":
text = item.get("text") or "" text = item.get("text") or ""
if not text: if not text:
return [] return []
return [{"kind": "plan", "id": item_id, "ts": ts, return [{"kind": "plan", "id": item_id, "ts": ts, "thread": thread,
"text": text[:300]}] "text": text[:300]}]
return [] return []
def _trace_file_path(state_path, task_name):
"""链路文件路径:state 文件同目录下的 host-traces/<清洗后名字>.jsonl。
与 Server.set_trace_file、cloud 的 _trace_file_name 同款清洗,
三边算出同一名字,delegate 的父链路标记才能落进父任务文件。
"""
safe = re.sub(r"[^A-Za-z0-9_.-]", "-", str(task_name))
return os.path.join(
os.path.dirname(os.path.abspath(state_path)),
"host-traces", safe + ".jsonl")
def _append_trace_marker(state_path, task_name, entry):
"""往另一任务的链路文件追加一条标记(delegate 向父链路报到用)。
尽力而为:目录建不出、写失败都不影响回合本身。父链路的
行数上限由属主进程自己的计数器管,这里不重复计数(标记
一轮至多两条,超限风险可忽略)。
"""
try:
path = _trace_file_path(state_path, task_name)
os.makedirs(os.path.dirname(path), exist_ok=True)
with open(path, "a", encoding="utf-8") as f:
f.write(json.dumps(entry, ensure_ascii=False) + "\n")
except Exception:
pass
class Server: class Server:
"""单个 app-server 进程,多 thread 并发。""" """单个 app-server 进程,多 thread 并发。"""
@@ -302,12 +336,8 @@ class Server:
时 trace_path 保持 None,不影响回合。 时 trace_path 保持 None,不影响回合。
""" """
try: try:
safe = re.sub(r"[^A-Za-z0-9_.-]", "-", str(task_name)) self.trace_path = _trace_file_path(state_path, task_name)
trace_dir = os.path.join( os.makedirs(os.path.dirname(self.trace_path), exist_ok=True)
os.path.dirname(os.path.abspath(state_path)),
"host-traces")
os.makedirs(trace_dir, exist_ok=True)
self.trace_path = os.path.join(trace_dir, safe + ".jsonl")
self._trace_lines = 0 self._trace_lines = 0
except Exception: except Exception:
self.trace_path = None self.trace_path = None
@@ -2566,11 +2596,26 @@ def main():
"created_at": time.time(), "turns": []} "created_at": time.time(), "turns": []}
state["tasks"][args.name] = sub state["tasks"][args.name] = sub
save_state(args.state, state) save_state(args.state, state)
# 子代理链路:向父任务链路文件报到(开始 + 结束各一条),
# 服务端凭标记把子链路文件拼进父任务的代理链;子任务自己
# 的过程条目仍写它自己的链路文件(main 起动时按 --name 定)。
child_trace = (os.path.basename(srv.trace_path)
if srv.trace_path else None)
_append_trace_marker(
args.state, args.parent_task,
{"kind": "agent-start", "id": "agent:" + args.name,
"name": args.name, "prompt": (args.task or "")[:300],
"trace": child_trace, "ts": int(time.time() * 1000)})
result = run_task_turn(srv, state, sub, args.task, result = run_task_turn(srv, state, sub, args.task,
args.model, args.effort, args.sandbox, args.model, args.effort, args.sandbox,
timeout=args.turn_timeout, timeout=args.turn_timeout,
interrupt_file=_interrupt_request_path(args.state, args.name)) interrupt_file=_interrupt_request_path(args.state, args.name))
save_state(args.state, state) save_state(args.state, state)
_append_trace_marker(
args.state, args.parent_task,
{"kind": "agent-done", "id": "agent:" + args.name + ":done",
"name": args.name, "status": sub["status"],
"trace": child_trace, "ts": int(time.time() * 1000)})
print(json.dumps({"sub_task": args.name, "status": sub["status"], print(json.dumps({"sub_task": args.name, "status": sub["status"],
"text": result.get("text", "")}, "text": result.get("text", "")},
indent=2, ensure_ascii=False)) indent=2, ensure_ascii=False))
+43
View File
@@ -6,6 +6,7 @@
import json import json
import os import os
import shutil
import sys import sys
import tempfile import tempfile
import unittest import unittest
@@ -130,6 +131,48 @@ class TraceFileTestCase(unittest.TestCase):
self.assertEqual(len(lines), 1) self.assertEqual(len(lines), 1)
self.assertEqual(json.loads(lines[0])["id"], "c1") self.assertEqual(json.loads(lines[0])["id"], "c1")
def test_summarize_carries_thread(self):
# 模型内子代理与主任务同文件时,服务端凭 thread 归属条目。
entries = _summarize_trace_item(
"item/completed",
{"item": {"type": "commandExecution", "id": "c1",
"command": "ls"},
"threadId": "th-sub", "completedAtMs": 1700000000000})
self.assertEqual(entries[0]["thread"], "th-sub")
entries = _summarize_trace_item(
"item/completed",
{"item": {"type": "commandExecution", "id": "c2",
"command": "ls"},
"completedAtMs": 1700000000000})
self.assertIsNone(entries[0]["thread"])
def test_append_trace_marker_to_parent_file(self):
# delegate 向父链路报到:路径清洗与 set_trace_file 一致,
# 写失败不抛(尽力而为,不影响回合)。
from host import _append_trace_marker, _trace_file_path
with tempfile.TemporaryDirectory() as d:
state_path = os.path.join(d, "host-state.json")
self.assertEqual(
_trace_file_path(state_path, "abc/def?id=1"),
os.path.join(d, "host-traces", "abc-def-id-1.jsonl"))
_append_trace_marker(
state_path, "parent task!",
{"kind": "agent-start", "id": "agent:rev",
"name": "rev", "prompt": "review"})
path = os.path.join(d, "host-traces", "parent-task-.jsonl")
with open(path, encoding="utf-8") as f:
entry = json.loads(f.read().strip())
self.assertEqual(entry["kind"], "agent-start")
self.assertEqual(entry["name"], "rev")
# 目录被文件占住时也不抛。
blocker = os.path.join(d, "host-traces")
shutil.rmtree(blocker)
with open(blocker, "w", encoding="utf-8") as f:
f.write("x")
_append_trace_marker(state_path, "parent task!",
{"kind": "agent-done", "id": "agent:rev:done",
"name": "rev", "status": "completed"})
if __name__ == "__main__": if __name__ == "__main__":
unittest.main() unittest.main()