Skip to content

🐛 fix(heterogeneous-agents): stream subagent Thread + fix parallel-tool orphan#14024

Merged
arvinxx merged 3 commits into
canaryfrom
fix/lobe-7365-cc-monitor-parentid-chain
Apr 21, 2026
Merged

🐛 fix(heterogeneous-agents): stream subagent Thread + fix parallel-tool orphan#14024
arvinxx merged 3 commits into
canaryfrom
fix/lobe-7365-cc-monitor-parentid-chain

Conversation

@arvinxx

@arvinxx arvinxx commented Apr 21, 2026

Copy link
Copy Markdown
Member

Summary

Three coordinated commits addressing parallel-tool orphan, subagent thread streaming, and a buffer-flush durability concern surfaced during code review.

  • Main-bucket isolation: subagent-tagged stream_chunk events are no longer forwarded to the main-agent gateway handler. Under a parallel main tool_use ([Grep, Agent]), subagent inner tools no longer overwrite main.tools[] in memory, so main's Task/Agent tool_use stays paired with its role:'tool' message (orphan tool-call banner no longer triggers).
  • Thread-bucket streaming via per-spawn sub-operation: each subagent spawn opens a child operation with parentOperationId = main op and context = { ...mainContext, threadId, scope: 'thread' }. The thread-scoped dispatcher binds to that sub-op's id, so every Thread bucket dispatch resolves through the standard internal_getConversationContextmessageMapKey path — no special-cased threadId override at the dispatch boundary. Cancel cascade + cleanup ride on the existing parent/child operation linkage.
  • Durable subagent buffer flush: finalizeSubagentRun only drains its accumulatedContent/accumulatedReasoning after messageService.updateMessage confirms the write. Failure paths now keep the buffer AND pin the flush target on SubagentRunState.pendingFlushTarget so the onComplete fallback retries against the streaming turn's assistant — never against the terminal row that the resultContent branch advanced currentAssistantMsgId onto.

The dispatcher streams Thread-bucket createMessage (user/assistant/tool seeds) and updateMessage (tools[]/content/reasoning/tool-result) on every chunk, so the Thread view streams token-by-token with the same cadence as the main bubble. Without it the Thread stayed empty until SWR re-fetched on next open (fetchAndReplaceMessages is main-topic scoped).

Test plan

  • Unit: heterogeneousAgentExecutor.test.ts 40/40 pass, incl. three regressions:
    • does NOT forward subagent-tagged stream_chunks to the gateway handler — main-bucket isolation under parallel main+subagent tool use.
    • streams subagent create/update dispatches via a thread-scoped sub-operation — verifies (1) startOperation is called with type: 'subagentThread', parented to the main op, with the Thread's ConversationContext; (2) every Thread bucket createMessage/updateMessage carries the sub-op id; (3) tool result lands on the in-thread tool message id; (4) no main-bucket leak; (5) completeOperation(subOpId) fires on finalize.
    • retains subagent buffers + pinned target when the finalize flush fails — stubs updateMessage to throw once on the streaming write, asserts the leftover content reaches DB across ≥2 attempts and every attempt targets the streaming turn's assistant — not the terminal row.
  • Type check clean (bun run type-check).
  • Message-slice tests 140/140 pass — internal_dispatchMessage is back to its original { operationId? } signature, no leaky parallel context path remains in the store boundary.
  • Local Electron repro (CC provider, two parallel runs of [Grep, Agent]):
    • Main bucket: both assistants keep tools = [Grep, Agent]; all 4 role:'tool' rows paired to their parent assistant; orphan DOM node count = 0.
    • Thread bucket populates in real time (14 rows: user + 2 subagent assistants with [Bash, Glob] then [Read×8] + 10 tool results); Thread view renders user→assistant(tools + streaming content).
    • Integrity checks across both runs: mainOrphans: [], threadOrphans: [], threadIntoMainBleed: [].

🤖 Generated with Claude Code

…ool orphan

When a main-agent step emits a parallel tool_use (e.g. `[Grep, Agent]`),
the gateway handler's stream_chunk branch was forwarding the subagent's
inner `tools_calling` chunks onto `currentAssistantMessageId` (main),
overwriting main.tools[] with subagent tools — main's own Task/Agent
tool_use then had no matching entry and every tool message under it
rendered with the "orphan tool call" banner.

Two coordinated changes:

1. Main-bucket isolation: the executor now drops subagent-tagged
   `stream_chunk` events before forwarding to the gateway handler. DB
   persistence continues via `persistSubagent*Chunk` so the subagent
   content is never lost; only the main-handler in-memory dispatch is
   suppressed for subagent chunks.

2. Thread-bucket streaming: `internal_dispatchMessage` now accepts a
   `threadId` override that snaps scope to `thread`, routing
   create/update payloads to the thread's `messagesMap` bucket. Each
   `SubagentRunState` carries a thread-scoped dispatcher; ensureSubagentRun
   seeds user + assistant on lazy Thread creation and at turn boundaries,
   persistToolBatch gets an `onToolCreated` hook that the subagent path
   uses to seed role:'tool' rows, persistSubagent*Chunk dispatches
   tools[] / content / reasoning updates on every chunk, and the
   tool_result branch mirrors subagent tool_result content (+ pluginState)
   into the thread bucket. Thread view now streams token-by-token with
   the same cadence as the main bubble.

Tests:
- `does NOT forward subagent-tagged stream_chunks to the gateway handler`
  — asserts main bucket isolation under parallel main+subagent tool use.
- `streams subagent create/update dispatches into the thread messagesMap
  bucket` — asserts user/assistant/tool createMessage dispatches land in
  the thread scope, plus streaming updateMessage for tools[], content,
  and tool_result, with no bleed into the main bucket.

Local repro verified end-to-end: main assistant.tools=[Grep, Agent]
stays intact across two parallel runs, thread bucket populates 14 rows
(user + 2 subagent assistants with Bash/Glob then Read×8 + 10 tool
results) during the run, `mainOrphans`/`threadOrphans`/
`threadIntoMainBleed` all empty, orphan warning DOM count = 0.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
@vercel

vercel Bot commented Apr 21, 2026

Copy link
Copy Markdown

The latest updates on your projects. Learn more about Vercel for GitHub.

Project Deployment Actions Updated (UTC)
lobehub Ready Ready Preview, Comment Apr 21, 2026 2:04pm

Request Review

@sourcery-ai sourcery-ai Bot left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We've reviewed this pull request using the Sourcery rules engine

@codecov

codecov Bot commented Apr 21, 2026

Copy link
Copy Markdown

Codecov Report

❌ Patch coverage is 95.26316% with 9 lines in your changes missing coverage. Please review.
✅ Project coverage is 66.91%. Comparing base (b02b727) to head (a086d65).
⚠️ Report is 2 commits behind head on canary.

Additional details and impacted files
@@            Coverage Diff             @@
##           canary   #14024      +/-   ##
==========================================
+ Coverage   66.86%   66.91%   +0.05%     
==========================================
  Files        2100     2103       +3     
  Lines      179535   179787     +252     
  Branches    21192    17778    -3414     
==========================================
+ Hits       120038   120299     +261     
+ Misses      59373    59364       -9     
  Partials      124      124              
Flag Coverage Δ
app 59.57% <95.26%> (+0.08%) ⬆️
database 92.27% <ø> (ø)
packages/agent-runtime 79.72% <ø> (ø)
packages/context-engine 83.18% <ø> (ø)
packages/conversation-flow 92.40% <ø> (ø)
packages/file-loaders 87.02% <ø> (ø)
packages/memory-user-memory 74.74% <ø> (ø)
packages/model-bank 99.86% <ø> (ø)
packages/model-runtime 84.22% <ø> (ø)
packages/prompts 69.08% <ø> (ø)
packages/python-interpreter 92.90% <ø> (ø)
packages/ssrf-safe-fetch 0.00% <ø> (ø)
packages/utils 87.95% <ø> (ø)
packages/web-crawler 88.66% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

Components Coverage Δ
Store 66.55% <95.26%> (+0.17%) ⬆️
Services 51.71% <ø> (ø)
Server 66.77% <ø> (+<0.01%) ⬆️
Libs 52.57% <ø> (ø)
Utils 80.59% <ø> (ø)
🚀 New features to boost your workflow:
  • ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
  • 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

@chatgpt-codex-connector chatgpt-codex-connector Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 63c00290a2

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

  • Open a pull request for review
  • Mark a draft as ready
  • Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

Comment thread src/store/chat/slices/aiChat/actions/heterogeneousAgentExecutor.ts Outdated
…r-spawn sub-operation

Replace the threadId-override on `internal_dispatchMessage` with a
proper per-spawn child operation, eliminating the second context
expression at the dispatch boundary.

The previous design accepted `{ operationId, threadId? }` and snapped
scope to `'thread'` when the override was present. That was a leaky
parallel path to the operation registry — the same "which messagesMap
bucket should this dispatch hit?" question got answered two different
ways. `startOperation` already supports `parentOperationId` + context
inheritance + recursive cancel cascade, so the right move is to model
the subagent run as a first-class child op and let
`internal_getConversationContext` do its normal job.

Changes:
- Add `'subagentThread'` to `OperationType` (NOT in
  `AI_RUNTIME_OPERATION_TYPES` — it's a context container, not an
  independent loading state, so it shouldn't double-count for spinners).
- `executeHeterogeneousAgent` opens the sub-op in `beginSubagentRun`
  via `startOperation({ type: 'subagentThread', parentOperationId,
  context: { ...context, threadId, scope: 'thread' } })` and binds a
  thread-scoped dispatcher to that sub-op's id.
- `SubagentRunState.subOperationId` carries the id so `finalizeSubagentRun`
  can mark it completed when the spawn's tool_result arrives (or on the
  `onComplete` fallback for crash/abort paths). Cancel cascade + cleanup
  flow through the existing parent/child op linkage.
- Revert the `threadId` override in `internal_dispatchMessage` — the
  store boundary is back to a single context expression
  (`{ operationId? }`).

Test:
- Add `startOperation` mock to `createMockStore` (returns monotonic
  `sub-op-N` ids).
- Update the streaming regression to identify the sub-op via the
  `startOperation` call with `type: 'subagentThread'`, assert the
  sub-op's parent + context shape, filter Thread bucket dispatches by
  `ctx.operationId === subOperationId`, and verify
  `completeOperation(subOperationId)` fires when the run finalizes.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…sh confirms

`finalizeSubagentRun`'s buffer reset used to run unconditionally after
the flush try/catch, so a transient `messageService.updateMessage`
failure silently wiped the accumulated streamed text/reasoning — the
later `onComplete` fallback then had nothing left to retry, leaving the
subagent's streamed content absent from persisted thread history.

Move the clear into the success branch. A second concern surfaces once
the clear moves: after the flush block, the `resultContent` branch
advances `currentAssistantMsgId` to the newly created terminal
assistant, so a naive retry that reads `currentAssistantMsgId` would
overwrite the authoritative terminal content with the leftover streamed
buffer — corrupting the subagent summary with stale partial text.

Pin the flush target via a new `SubagentRunState.pendingFlushTarget`:
captured before the DB attempt, carried on the run when the flush
fails, cleared alongside the buffers on success. The retry uses the
pinned target instead of the live `currentAssistantMsgId`, so leftover
streamed buffers always land on the streaming turn's assistant — never
on the terminal row.

Test: `retains subagent buffers + pinned target when the finalize flush
fails` stubs `updateMessage` to throw once for the subagent streaming
write, runs streamed text → spawn `tool_result` → `onComplete`, and
asserts (1) the leftover content eventually reaches DB across ≥2
write attempts and (2) every attempt targets the streaming turn's
assistant — not the terminal row created by `resultContent`.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
@arvinxx arvinxx changed the title ✨ feat(heterogeneous-agents): stream subagent Thread + fix parallel-tool orphan 🐛 fix(heterogeneous-agents): stream subagent Thread + fix parallel-tool orphan Apr 21, 2026
@arvinxx arvinxx merged commit f3fca50 into canary Apr 21, 2026
34 checks passed
@arvinxx arvinxx deleted the fix/lobe-7365-cc-monitor-parentid-chain branch April 21, 2026 14:18
@arvinxx arvinxx mentioned this pull request Apr 27, 2026
arvinxx added a commit that referenced this pull request Apr 27, 2026
# 🚀 LobeHub v2.1.53 (20260427)

**Release Date:** April 27, 2026
**Since v2.1.52:** 194 merged PRs · 17 contributors

> Introduce Heterogeneous Agent — Claude Code and Codex run as
first-class desktop runtimes, paired with a new Agent Signal package,
sharper desktop UX, and a wave of flagship model additions.

---

## ✨ Highlights

- **Introduce Heterogeneous Agent** — Claude Code and Codex run as
first-class desktop agents: subagent rendering, partial-message
streaming, multi-turn resume, terminal error surfacing, rich tool
inspectors, and runtime polish. (#14162, #13754, #14067, #14001, #13970,
#13942)
- **Screen capture & Quick Chat tray** — New desktop screen capture
overlay (macOS permission-gated) with Quick Chat tray and upload
pipeline improvements; chat input auto-focuses on overlay mount.
(#13818, #14097, #14105)
- **Desktop topic & tab UX** — Dedicated topic popup window with
cross-window sync, Cmd+W/Cmd+T tab shortcuts, TabBar polish, recent
working directories expanded to 20, and human approval notifications.
(#13957, #13983, #13972, #14036, #14092)
- **Git workflow built-in** — One-click pull/push from the branch chip,
ahead/behind badge, and submodule/worktree repo detection. (#14041,
#13980, #13978)
- **Agent Signal package** — New `@lobechat/agent-signal` runtime for
dynamic memory feedback signals, with OTel metrics and self-iteration in
Lab. (#14157, #14170, #14159, #14169, #14187)
- **New models** — Claude Opus 4.7 with `xhigh` effort tier, GPT-5.5,
DeepSeek V4 Flash/Pro with reasoning slider, Kimi K2.6, MiMo-V2.5/Pro,
gpt-image-2, Qwen3.6 Flash/Plus, and Pixverse-c1. (#13903, #14147,
#14114, #14004, #14089, #14039, #13923)
- **New providers** — OpenCode Zen, OpenCode Go, and Azure OpenAI Router
runtime. (#13943, #14064, #13823)
- **Mobile settings overhaul** — Full settings menu and responsive
profile layout for mobile. (#14019)

---

## 🏗️ Heterogeneous Agent

- Claude Code runtime, working-directory awareness, and sidebar polish.
(#13970)
- CC subagent rendering with persistent streamed text; parallel-tool
orphan fix. (#14001, #13968, #14024)
- Per-step usage persisted to each step assistant message. (#13964)
- Per-phase workflow expand defaults; full-expand toggle with
three-level expansion. (#14171, #13906)
- Hetero-mode actions bar; tool inspector polish. (#13963, #14034,
#14030)
- Codex desktop integration with rich tool rendering and devtools
preview. (#14067, #14100)
- Codex terminal error surfacing and CLI output tracing. (#14166)
- Tighten `isCanUseVision` default and add aggregator fallback. (#14172)
- Persist `ccSessionId` in topic metadata for CC multi-turn resume.
(#13902)
- CC account card, topic filter, and integration polish. (#13955,
#13942, #13950)
- Token-level deltas streamed via `--include-partial-messages`. (#13929)

---

## 🧠 Agent Signal & Self-Iteration

- New `@lobechat/agent-signal` package with dynamic feedback signals.
(#14157)
- AgentSignalRuntime wired through agent-tracing and observability-otel
metrics. (#14170, #14159)
- Self-iteration feature flag added to Lab; front-side flag check.
(#14169, #14186)
- Signal policy for receiving memory feedback dynamically. (#14187)

---

## 💬 Conversation

- Queue follow-up sends during running CC turns. (#14179)
- Persist per-topic chat scroll position; pin user message + fold long
messages. (#14191, #14056)
- Inline resend when editing last user message. (#14080)
- Disable first-block markdown streaming to prevent flicker. (#14193,
#13904)
- Prevent Markdown stream replay when vlist remounts streaming items.
(#14086)
- Stop repinning after manual scroll; unify scroll-to-user + spacer
hooks. (#14099, #14132)

---

## 📱 Platforms & Integrations

### Desktop / Electron

- Screen capture overlay, Quick Chat tray, and upload pipeline
improvements. (#13818)
- macOS permission gate for screen capture; auto-focus chat panel input.
(#14097, #14105)
- Dedicated topic popup window with cross-window sync. (#13957)
- TabBar polish: `+` button for new topic, dark theme blend, close icon
by default. (#13972, #14203, #13973)
- Recent working directories expanded from 5 to 20; submodule/worktree
repo detection. (#14036, #13978)
- Cmd+W / Cmd+T tab shortcuts and global shortcut consolidation.
(#13983, #13880)
- Linux icon configuration; human approval desktop notifications.
(#14042, #14092)

### Git Workflow

- One-click pull/push from branch chip; ahead/behind badge with
refactored GitCtr. (#14041, #13980)

### Mobile

- Full settings menu and responsive profile layout. (#14019)
- Agent route added to mobile router; mobile agent topic route
registered. (#14103, #14158)
- Session list skeleton row layout corrected. (#14040)

### Bot / Messaging

- DM strategy support; bot emoji and markdown render optimization.
(#14201, #14091, #14140)
- Slack webhook fix; bot platform setup guide reference. (#14052,
#14121)

---

## 🤖 Models & Providers

### New models

- **Claude Opus 4.7** with `xhigh` effort tier; strip temperature/top_p.
(#13903, #13909)
- **GPT-5.5**. (#14147)
- **DeepSeek V4** Flash/Pro cards with reasoning slider; cache-hit and
Pro discount pricing. (#14114, #14209, #14196, #14131)
- **Kimi K2.6** model with LobeHub-hosted card. (#14004, #14006)
- **MiMo-V2.5 / V2.5-Pro**. (#14089)
- **gpt-image-2**, **Qwen3.6 Flash/Plus**, **Pixverse-c1**. (#14039,
#13923)

### New providers

- **OpenCode Zen** and **OpenCode Go** with env-var support. (#13943,
#14064)
- **Azure OpenAI Router** runtime support. (#13823)
- Model alias mapping for image and video runtimes. (#13896)
- Seedance video models migrated to Dreamina. (#14144)

### Runtime reliability

- Sanitize invalid tool_call arguments to unbreak strict providers.
(#14033)
- Tolerate null `function.name` in streaming tool_call deltas. (#14139)
- Preserve Gemini 3 `thoughtSignature` in `call_tools_batch`
normalization. (#14032)
- Downgrade `image_url` parts when target model lacks vision. (#14029)
- Preserve Cloudflare provider error context. (#14136)
- Use `safety_identifier` for OpenAI Responses API. (#14148)
- Unwrap underlying PG error in `formatErrorEventData`. (#14038)

---

## 🖥️ User Experience

- **Onboarding** — Preset agent naming suggestions, structured hunk ops
for `updateDocument`, persona analytics snapshot, footer promotion
pipeline, wrap-up button. (#13931, #13989, #13930, #13853, #13934)
- **Document workflow** — Agent documents promoted as primary workspace
panel; history management and compare workflow; web-crawl docs
associated with agent documents. (#13924, #13725, #13893)
- **cmdk** — Agent identity surfaced on topic search results;
topic/message search scoped to current agent. (#14204, #13960)
- **Floating chat panel** and workspace improvements. (#13887)
- **Topic completion status** with dropdown action and filter. (#14005)

---

## 🔧 Tooling

- Redis-backed feature flag provider for runtime config. (#14098)
- Vite upgraded to 8.0.0 with Rolldown strict execution order. (#12720,
#14058)
- `@lobechat/model-bank` automated npm release with provenance. (#14015,
#14017, #14018)
- Skill activation fallback when `activateTools` cannot find identifier.
(#14010)
- Cron tool: timezone and existing jobs injected into system prompt;
clarified `lobe-gtd` and `lobe-cron` descriptions. (#14012, #14013)

---

## 🔒 Security & Reliability

- **Security:** uuid bumped to v14 (advisory). (#14083)
- **Security:** validate avatar URL and scope old-avatar deletion to
owner. (#13982)
- **Security:** clear OIDC sessions on better-auth signout; return 401
(not 500) for expired OIDC JWT. (#13916, #14014)
- **Reliability:** scope pending-approval check to current assistant
turn. (#14182)
- **Reliability:** sanitize heterogeneous-agent attachment cache
filenames. (#13937)
- **Reliability:** reduce subagent task status error noise. (#14026)

---

## 👥 Contributors

Huge thanks to **17 contributors** who shipped **194 merged PRs** this
week.

@hardy · @shaun0927 · @hezhijie0327 · @sxjeru · @arvinxx · @Innei ·
@tjx666 · @lijian · @neko · @rdmclin2 · @AmAzing129 · @sudongyuer ·
@CanisMinor · @rivertwilight

Plus @lobehubbot and renovate[bot] for maintenance.

---

**Full Changelog**:
v2.1.52...v2.1.53
arvinxx added a commit that referenced this pull request May 15, 2026
…y handler (#14838)

The forwarding guard only filtered `stream_chunk` events. `tool_start` and
`tool_end` for subagent inner tools still reached the main handler, where
`tool_end` fired a `fetchAndReplaceMessages(main)` on every subagent inner
tool result — wasted work AND a state-drift window that surfaced as the
"orphan tool call" banner on the spawn's bubble even after DB had settled.

`tool_start(subagent)` was also leaking `dispatchOnBeforeCall` invocations
against the main context for what is actually a subagent inner tool, firing
renderer onBeforeCall hooks in the wrong scope.

Broadens the guard to drop ALL events with `event.data.subagent`. Safe
because:
- `tool_result(subagent)` is already handled inline at executor:1407 with
  an early `return`.
- `stream_chunk(subagent)` is routed through `persistSubagent*Chunk` into
  the per-spawn thread scope; the subagent's own in-thread renderer state
  is streamed via the thread-scoped dispatcher introduced in #14024.
- `tool_start` / `tool_end` are pure renderer-notification hooks; the
  subagent has no business firing them on the main bucket.

Regression test asserts:
- No forwarded event with `event.data.subagent` reaches the handler.
- Main's own `tool_start` / `tool_end` (no subagent flag) still reach
  the handler so the main bubble's animation + onAfterCall hooks fire.

Closes LOBE-8991.

Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant