Skip to content

Self-hosted workflows for coding-agent delivery. Orchestrate CodeBuddy, Codex and Claude Code with scripts and human approval gates. Includes Local workflows for coding assistants.

feat(chatbot): show the model a subagent ran on

roberttwang4d2ef214d1f3
diff --git a/CHANGELOG.md b/CHANGELOG.md
index ba99076f..3e690806 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -4,6 +4,7 @@ Platform releases use `vX.Y.Z` tags and publish `ghcr.io/tencent/loopforge`. Loo
## Unreleased
+- A conversation's 子 Agent shows the model it ran on: its row in the artifacts list reads 「已完成 · 模型 …」, the header of an opened subagent names the model next to its status, and each dispatch in 会话过程 notes 「模型 · …」 under the task it was given. The model is the one the Agent call named for the subagent (the `model` argument, for example `lite` or a model ID), else the model of the reply that dispatched it, which the subagent inherits; a dispatch still running uses the model the running reply was sent with. Activity events of an Agent call carry the new `model` field only when the call named one; replies saved before this change show the model of their reply.
- The built-in backend, frontend and defect-fix SOPs gain a 「发起 MR」 node between 人工验收 and 交付汇总 (#717). It commits the work branch, pushes it and opens an MR/PR against the run's target branch, then records the link, branches and commit SHA in `mr-body.md`. Its command gate checks every writable repository of the run: one with commits beyond the target branch fails while it has uncommitted changes, while its checkout is off the run's work branch, or until `git ls-remote origin` shows that work branch at HEAD. A repository the run left unchanged is skipped. A second gate of the new mode `mr` has the engine look the MR up on the host with the same lookup delivery tracking uses (GitHub and Gongfeng): every writable repository whose work branch on origin carries commits beyond the target branch needs an MR from that work branch into that target branch that carries the pushed tip: an open one whose head is the tip (or whose host reports no head), or a merged one whose head is the tip. An agent that reports pass without opening an MR, opens it against the wrong branch, or leaves only an earlier merged MR on a reused work branch does not reach 交付汇总. The lookup runs as the run: on GitHub with the Git token the run pushed with (its responsible user's), falling back to the Owner of the run's own project; on Gongfeng through the run project's MCP as the responsible user. A repository bound to several projects is no longer checked as whichever project bound it first, and a failed lookup names the identity it used. A host the platform cannot query (an SSH repository, a failed lookup) goes straight to a person. `mr` gates take no commands, and the role editor keeps them when a role is saved. The gates retry once with their feedback and then wait for a person, so a run no longer succeeds with every source change left in a workspace that is reclaimed right after success. Run-scoped commands now get `AGENTFLOW_WORK_BRANCHES` (one `dir<TAB>work branch<TAB>target branch` line per writable repository) and repository-scoped commands get `AGENTFLOW_WORK_BRANCH` and `AGENTFLOW_BASE_BRANCH`; the target branch is the one frozen into the run, including a requirement's release branch and a scheduled SOP's integration branch. 交付汇总 no longer pushes or opens a PR itself. Runs without a bound repository still deliver from the task workspace. New installs get the new templates from the seed. Existing installs keep their templates until an operator runs `migrations/033_generic_sop_mr_node.sql` by hand (see docs/deployment.md); it rewrites a template only while its node graph and 交付汇总 position are the shipped original and its 交付汇总 prompt only while unedited. SOPs already adopted into a project are copies and keep their old graph. The product SOP delivers documents only and is unchanged.
- A conversation reply that is still running keeps running through a slow heartbeat. A poll, the conversation list, a follow-up message or a steer used to record it as interrupted after 45 seconds without a stored heartbeat; the process then cancelled the CLI and the next message failed with 「工作区正在使用或保存,请稍后重试」 for up to 15 minutes. Reads of a turn this process executes no longer reap it, and an unchanged heartbeat on MySQL no longer counts as a lost turn. On MySQL a heartbeat's age is measured on the database clock that stamped it, so skew between an API host and MySQL neither reaps a live turn nor keeps a dead one; whether a lease holder stopped renewing is likewise judged from `run_workspace_checkpoints.lease_renewed_at`, stamped on the database clock, not from another replica's `lease_until`. When a turn whose process died is reaped, the workspace lease it recorded is released if nothing renewed it for 45 seconds, so the next message starts at once; a lease another turn took, one still renewed on another replica, and a background save are kept. `migrations/032_assistant_generation_workspace_lease.sql` adds the three columns by hand; startup AutoMigrate adds them too.
diff --git a/chatbot/src/ArtifactPanel.jsx b/chatbot/src/ArtifactPanel.jsx
index 70d3ec3a..6681a1b2 100644
--- a/chatbot/src/ArtifactPanel.jsx
+++ b/chatbot/src/ArtifactPanel.jsx
@@ -206,6 +206,7 @@ export default function ArtifactPanel({
const subagents = collectSubagents(
chat?.session?.messages || [],
live ? chat?.live?.activity || [] : [],
+ chat?.model || "",
);
// A new subagent is keyed by its dispatch until its agent ID arrives, so
// an opened one is found by either.
@@ -344,6 +345,7 @@ export default function ArtifactPanel({
subagentName(agent),
agent.subagent,
agent.agentId,
+ agent.model,
subagentStatus(agent, live).label,
...agent.dispatches.map((step) => step.detail),
"子 Agent",
@@ -530,8 +532,8 @@ export default function ArtifactPanel({
{selectedAgent ? (
<>
<span className="artifact-node-title">{subagentName(selectedAgent)}</span>
- <small className="node-header-meta" title={[selectedAgent.subagent, selectedAgent.agentId].filter(Boolean).join(" · ")}>
- {subagentStatus(selectedAgent, live).label}
+ <small className="node-header-meta" title={[selectedAgent.subagent, selectedAgent.agentId, selectedAgent.model].filter(Boolean).join(" · ")}>
+ {[subagentStatus(selectedAgent, live).label, selectedAgent.model].filter(Boolean).join(" · ")}
</small>
</>
) : null}
@@ -776,6 +778,7 @@ export default function ArtifactPanel({
<small>
<i className={`artifact-run-dot status-${{ done: "succeeded", running: "running", failed: "failed" }[status.tone] || "canceled"}`} />
{status.label}
+ {agent.model ? ` · 模型 ${agent.model}` : ""}
{count > 1 ? ` · 派发 ${count} 次` : ""}
{agent.dispatches.at(-1)?.detail
? ` · ${agent.dispatches.at(-1).detail}`
diff --git a/chatbot/src/SubagentSession.jsx b/chatbot/src/SubagentSession.jsx
index 642e7af5..b3e42a34 100644
--- a/chatbot/src/SubagentSession.jsx
+++ b/chatbot/src/SubagentSession.jsx
@@ -34,8 +34,8 @@ function processSteps(step) {
}
// SubagentSession shows one subagent like an SOP node: what Main Bot asked,
-// how it worked and what it reported, once per dispatch when it was resumed,
-// plus its collected deliverables and what it published.
+// the model it ran on, how it worked and what it reported, once per dispatch
+// when it was resumed, plus its collected deliverables and what it published.
export function SubagentSession({ agent, live = false, artifacts = [], onOpenArtifact }) {
const [tab, setTab] = useState("conversation");
useEffect(() => setTab("conversation"), [agent.key]);
@@ -96,6 +96,11 @@ export function SubagentSession({ agent, live = false, artifacts = [], onOpenArt
<RichContent message={step.input || step.detail} />
</article>
) : null}
+ {agent.models?.[step.id] ? (
+ <p className="muted subagent-model" aria-label="运行模型">
+ 模型 · {agent.models[step.id]}
+ </p>
+ ) : null}
{resumeMissed(step) ? (
<p className="muted subagent-resume-missed" role="note">
复用未生效:没有接着 {step.resumed} 做,已新开子 Agent。
diff --git a/chatbot/src/activity.js b/chatbot/src/activity.js
index 0bf97b64..c107e082 100644
--- a/chatbot/src/activity.js
+++ b/chatbot/src/activity.js
@@ -104,6 +104,7 @@ export function mergeActivity(previous = [], events = []) {
if (event.subagent) existing.subagent = event.subagent;
if (event.role) existing.role = event.role;
if (event.input) existing.input = event.input;
+ if (event.model) existing.model = event.model;
// A resumed dispatch names its agent before it finishes. Keep the
// target apart: the CLI may still finish it under a new agent.
if (event.agentId) existing.agentId = existing.resumed = event.agentId;
@@ -118,6 +119,7 @@ export function mergeActivity(previous = [], events = []) {
...(event.subagent ? { subagent: event.subagent } : {}),
...(event.role ? { role: event.role } : {}),
...(event.input ? { input: event.input } : {}),
+ ...(event.model ? { model: event.model } : {}),
...(event.agentId ? { agentId: event.agentId, resumed: event.agentId } : {}),
});
}
@@ -276,20 +278,27 @@ export const resumeMissed = (step) =>
// collectSubagents lists every subagent of a conversation, newest last. A
// dispatch that resumed an earlier agent joins it, so one entry holds every
// task that agent was given.
-export function collectSubagents(messages = [], liveSteps = []) {
+// A subagent runs on the model its dispatch named, else on the model of the
+// reply that dispatched it; a live dispatch has no saved reply yet and uses
+// liveModel, the model the running reply was sent with. agent.model is the
+// latest dispatch's, agent.models[step.id] each dispatch's.
+export function collectSubagents(messages = [], liveSteps = [], liveModel = "") {
const agents = [];
// messageIndex is the saved reply that holds the dispatch; a live one has
// none yet, and its collected files are not attachments until it is saved.
- const add = (step, messageIndex = null) => {
+ const add = (step, messageIndex = null, replyModel = "") => {
const key = step.agentId || step.id;
let agent = agents.find((item) => item.key === key || (step.agentId && item.agentId === step.agentId));
if (!agent) {
- agent = { key, agentId: step.agentId || "", role: "", subagent: "", dispatches: [], published: [] };
+ agent = { key, agentId: step.agentId || "", role: "", subagent: "", model: "", models: {}, dispatches: [], published: [] };
agents.push(agent);
}
if (step.agentId) agent.agentId = step.agentId;
agent.role = step.role || agent.role;
agent.subagent = step.subagent || agent.subagent;
+ const model = String(step.model || replyModel || "").trim();
+ if (model) agent.models[step.id] = model;
+ agent.model = model || agent.model;
agent.dispatches.push(step);
// What it published, and the role deliverables the platform collected
// when it finished (a required one it did not write is marked missing).
@@ -309,9 +318,9 @@ export function collectSubagents(messages = [], liveSteps = []) {
if (message.role !== "assistant") return;
savedActivity(message)
.filter(isDispatch)
- .forEach((step) => add(step, index));
+ .forEach((step) => add(step, index, message.model));
});
- liveSteps.filter(isDispatch).forEach((step) => add(step));
+ liveSteps.filter(isDispatch).forEach((step) => add(step, null, liveModel));
return agents;
}
// While streaming, the latest paragraph is the answer only once nothing has
diff --git a/chatbot/src/styles.css b/chatbot/src/styles.css
index 4637eabd..622723ed 100644
--- a/chatbot/src/styles.css
+++ b/chatbot/src/styles.css
@@ -1263,6 +1263,12 @@ small {
.node-conversation-view .node-message:first-child {
margin-top: 20px;
}
+/* The model a subagent dispatch ran on, under the task it was given. */
+.subagent-session .subagent-model {
+ margin: 6px 0 0;
+ font-size: 12px;
+ overflow-wrap: anywhere;
+}
.composer-toolbar {
display: flex;
align-items: center;
diff --git a/chatbot/tests/subagents.test.jsx b/chatbot/tests/subagents.test.jsx
index b69a4821..823e8392 100644
--- a/chatbot/tests/subagents.test.jsx
+++ b/chatbot/tests/subagents.test.jsx
@@ -82,6 +82,57 @@ describe("dispatched subagents", () => {
await userEvent.click(screen.getByRole("button", { name: /测试报告/ }));
expect(setSelected).toHaveBeenLastCalledWith(report);
});
+
+ test("a subagent runs on the model its dispatch named, else on its reply's", () => {
+ const dispatch = (id, extra = {}) => ({ kind: "tool", id, tool: "Agent", detail: id, subagent: "role-qa", ...extra, at });
+ const steps = mergeActivity([], events(dispatch("c0"), dispatch("c0", { model: "lite" })));
+ expect(steps[0].model).toBe("lite");
+ const agents = collectSubagents(
+ [
+ { role: "assistant", model: "claude-opus-4-5", activity: [dispatch("c1"), { kind: "toolDone", id: "c1", agentId: "agent-aaa", at }] },
+ { role: "assistant", model: "claude-opus-4-5", activity: [dispatch("c2", { model: "claude-haiku-4-5", agentId: "agent-aaa" }), { kind: "toolDone", id: "c2", agentId: "agent-aaa", at }] },
+ { role: "assistant", activity: [dispatch("c3"), { kind: "toolDone", id: "c3", agentId: "agent-ccc", at }] },
+ ],
+ [dispatch("c4")],
+ "gpt-5",
+ );
+ expect(agents.map((agent) => [agent.key, agent.model, agent.models])).toEqual([
+ ["agent-aaa", "claude-haiku-4-5", { c1: "claude-opus-4-5", c2: "claude-haiku-4-5" }],
+ // A reply saved without a model reports none.
+ ["agent-ccc", "", {}],
+ // A live dispatch runs on the model the running reply was sent with.
+ ["c4", "gpt-5", { c4: "gpt-5" }],
+ ]);
+ });
+
+ test("the 子 Agent card and its session show the model it ran on", async () => {
+ const messages = [
+ { role: "user", content: "补单测" },
+ {
+ role: "assistant",
+ model: "claude-opus-4-5",
+ activity: [
+ { kind: "tool", id: "c1", tool: "Agent", detail: "补单测", input: "给 a.go 补单测", role: "测试工程师", at },
+ { kind: "toolDone", id: "c1", agentId: "agent-aaa", output: "用例都通过了", at },
+ ],
+ },
+ ];
+ const chat = { session: { id: "s1", projectId: "p1", messages }, projectId: "p1", projects: [] };
+ let selected = null;
+ const setSelected = vi.fn((value) => (selected = value));
+ const panel = () => (
+ <div className="chat-shell">
+ <ArtifactPanel artifacts={[]} runs={[]} selected={selected} setSelected={setSelected} chat={chat} onClose={vi.fn()} open />
+ </div>
+ );
+ const view = render(panel());
+ const card = screen.getByRole("button", { name: /测试工程师/ });
+ expect(card.textContent).toContain("已完成 · 模型 claude-opus-4-5");
+ await userEvent.click(card);
+ view.rerender(panel());
+ expect(screen.getByLabelText("运行模型").textContent).toBe("模型 · claude-opus-4-5");
+ expect(screen.getByText("已完成 · claude-opus-4-5")).toBeTruthy();
+ });
});
describe("role deliverables", () => {
diff --git a/internal/domain/assistant.go b/internal/domain/assistant.go
index dff9d0eb..49638af6 100644
--- a/internal/domain/assistant.go
+++ b/internal/domain/assistant.go
@@ -35,6 +35,10 @@ type AssistantActivityEvent struct {
Subagent string `json:"subagent,omitempty"`
Role string `json:"role,omitempty"`
AgentID string `json:"agentId,omitempty"`
+ // Model is the model an Agent dispatch asked its subagent to run on. It
+ // is empty when the subagent inherits the model of the reply that
+ // dispatched it.
+ Model string `json:"model,omitempty"`
// Input is the task an Agent dispatch hands the subagent and Output the
// report it returns; the side panel shows them like an SOP node session.
Input string `json:"input,omitempty"`
diff --git a/internal/httpapi/assistant_stream.go b/internal/httpapi/assistant_stream.go
index 41ffe0a4..ad30600e 100644
--- a/internal/httpapi/assistant_stream.go
+++ b/internal/httpapi/assistant_stream.go
@@ -78,10 +78,10 @@ func (f *assistantStreamFilter) acceptTool(ev assistantStreamEvent) bool {
}
// The CLI announces a call with empty input and repeats it once
// the input is complete. Keep the repeat only when it adds
- // something: the detail, or an Agent call's type and task.
- seen := ev.Detail + "\x00" + ev.Subagent + "\x00" + ev.Input + "\x00" + ev.AgentID
+ // something: the detail, or an Agent call's type, task and model.
+ seen := ev.Detail + "\x00" + ev.Subagent + "\x00" + ev.Input + "\x00" + ev.AgentID + "\x00" + ev.Model
if previous, ok := f.seenTools[ev.ID]; ok {
- if seen == previous || (ev.Detail == "" && ev.Subagent == "" && ev.Input == "") {
+ if seen == previous || (ev.Detail == "" && ev.Subagent == "" && ev.Input == "" && ev.Model == "") {
return false
}
}
@@ -293,10 +293,12 @@ func assistantToolStreamEvent(id, name string, input json.RawMessage) (assistant
SubagentType string `json:"subagent_type"`
Prompt string `json:"prompt"`
Resume string `json:"resume"`
+ Model string `json:"model"`
}
_ = json.Unmarshal(input, &dispatch)
event.Subagent = strings.TrimSpace(dispatch.SubagentType)
event.Input = truncateRunes(strings.TrimSpace(dispatch.Prompt), 8000)
+ event.Model = assistantDispatchModel(dispatch.Model)
// A resumed dispatch names its agent up front, so the panel can put
// it on that agent's row while it runs.
if m := assistantAgentIDRe.FindStringSubmatch("Agent ID: " + strings.TrimSpace(dispatch.Resume)); m != nil {
@@ -308,6 +310,17 @@ func assistantToolStreamEvent(id, name string, input json.RawMessage) (assistant
var assistantAgentIDRe = regexp.MustCompile(`Agent ID:\s*(agent-[A-Za-z0-9-]+)`)
+// assistantDispatchModel is the model an Agent call names for its subagent.
+// No model, or one that only says to inherit, leaves it to the reply's own.
+func assistantDispatchModel(model string) string {
+ model = strings.TrimSpace(model)
+ switch strings.ToLower(model) {
+ case "", "default", "inherit":
+ return ""
+ }
+ return truncateRunes(model, 128)
+}
+
// assistantResultText reads a tool result, which is either a string or a list
// of text blocks.
func assistantResultText(content json.RawMessage) string {
diff --git a/internal/httpapi/assistant_stream_filter_test.go b/internal/httpapi/assistant_stream_filter_test.go
index aafdb7e9..139c83a1 100644
--- a/internal/httpapi/assistant_stream_filter_test.go
+++ b/internal/httpapi/assistant_stream_filter_test.go
@@ -56,6 +56,8 @@ func TestAssistantStreamFilterAcceptToolDedupes(t *testing.T) {
{"identical repeat", assistantStreamEvent{Tool: "Bash", ID: "t1", Detail: "ls"}, false},
{"repeat with empty detail and input", assistantStreamEvent{Tool: "Bash", ID: "t1", AgentID: "a1"}, false},
{"agent call adds task", assistantStreamEvent{Tool: "Agent", ID: "t2", Subagent: "Explore", Input: "find x"}, true},
+ {"agent call adds model", assistantStreamEvent{Tool: "Agent", ID: "t2", Subagent: "Explore", Input: "find x", Model: "lite"}, true},
+ {"agent call repeats model", assistantStreamEvent{Tool: "Agent", ID: "t2", Subagent: "Explore", Input: "find x", Model: "lite"}, false},
{"no id empty tool", assistantStreamEvent{}, false},
{"no id same as last", assistantStreamEvent{Tool: "Agent", Detail: ""}, false},
{"no id new detail", assistantStreamEvent{Tool: "Agent", Detail: "y"}, true},
diff --git a/internal/httpapi/assistant_subagents_test.go b/internal/httpapi/assistant_subagents_test.go
index 1b6f3f9f..9c97acae 100644
--- a/internal/httpapi/assistant_subagents_test.go
+++ b/internal/httpapi/assistant_subagents_test.go
@@ -347,9 +347,21 @@ func TestEverySharedSkillIsInstalledHoweverManyAreShared(t *testing.T) {
func TestSubagentStreamEventsCarryTheirDispatch(t *testing.T) {
dispatch := assistantStreamEvents(`{"type":"assistant","message":{"content":[{"type":"tool_use","id":"call-1","name":"Agent","input":{"subagent_type":"role-qa","description":"补单测","prompt":"x"}}]}}`)
- if len(dispatch) != 1 || dispatch[0].Subagent != "role-qa" || dispatch[0].Parent != "" || dispatch[0].Input != "x" {
+ if len(dispatch) != 1 || dispatch[0].Subagent != "role-qa" || dispatch[0].Parent != "" || dispatch[0].Input != "x" || dispatch[0].Model != "" {
t.Fatalf("dispatch=%+v", dispatch)
}
+ // A dispatch that names a model keeps it; one that only inherits does not.
+ for input, want := range map[string]string{
+ `"model":"claude-sonnet-4-5"`: "claude-sonnet-4-5",
+ `"model":" lite "`: "lite",
+ `"model":"default"`: "",
+ `"model":"inherit"`: "",
+ } {
+ got := assistantStreamEvents(`{"type":"assistant","message":{"content":[{"type":"tool_use","id":"call-m","name":"Agent","input":{"subagent_type":"Explore","prompt":"x",` + input + `}}]}}`)
+ if len(got) != 1 || got[0].Model != want {
+ t.Fatalf("%s: dispatch=%+v, want model %q", input, got, want)
+ }
+ }
if got := labelRoleSubagent(dispatch[0], []roleSubagent{{Role: domain.AgentDefinition{ID: "qa", Name: "测试工程师"}}}); got.Role != "测试工程师" {
t.Fatalf("role label=%q", got.Role)
}