Error Invoking subagent cancel #296378
Description
Type: Bug
Subagents keep failing right when they are about to finish and create their .md document report. When this happens it completely ends the premium request instead of returning control back to the orchestrator agent.
Version: 1.109.4 (user setup)
Commit: c3a2684
Date: 2026-02-16T15:35:57.932Z
Electron: 39.3.0
ElectronBuildId: 13313481
Chromium: 142.0.7444.265
Node.js: 22.21.1
V8: 14.2.231.22-electron.0
OS: Windows_NT x64 10.0.26200
WSL
Summary:
While running an agent task with multiple sub-agents, execution stopped mid-validation. The Steering channel stopped sending messages. Using “Stop & Send” to recover appeared to undo or discard previously completed work (cleanup, decision docs, checklist updates). The UI shows “Error invoking subagent: Canceled,” and the agent never resumed or finalized.
Steps to Reproduce:
Start an agent task that launches multiple coordinated sub-agents (cleanup, decisions, checklist integration, validation).
Allow the agent to progress through initial batches and begin a full validation rerun (ruff, mypy, pytest).
Attempt to send a Steering message while the validation sub-agent is active.
Observe that Steering does not send.
Click “Stop & Send” to recover and send the message.
Expected Behavior:
Steering messages should be deliverable while a sub-agent is running, or the UI should clearly queue them.
If “Stop & Send” is used, previously completed agent work (files, docs, checklist state) should remain intact unless explicitly reverted, and the agent should resume or provide a consistent state.
Validation sub-agent should complete or fail with logs captured.
Actual Behavior:
Steering messages are not sent while the agent is in the validation phase.
“Stop & Send” results in apparent loss/reversion of prior work (cleanup and decision artifacts, checklist updates).
UI shows cancellation of the validation sub-agent without a clear recovery path.
No final consolidated state is produced.
Frequency: similar freezes are often. Not sure if all related
Impact:
Loss of agent progress and artifacts.
User uncertainty about workspace state and need to rerun tasks.
Reduced reliability of Steering for long-running agent jobs.
Workarounds Tried:
Waiting for completion (no progress).
Using “Stop & Send” (caused apparent reversion).
Often restore to previous checkpoint and try again.Some agents never seem to finish and I close the panel and start fresh.
Transcript / Visible Session Log (from UI):
Planning coordinated subagent tasks
I’m handling this with separate sub-agents in four non-overlapping streams: debris cleanup execution, ambiguous-file keep/revert decisioning, Trading-Win response integration into the checklist, and a full validation rerun. I’ll then consolidate what is fixed, what remains open, and exactly what is now confirmed by Trading-Win versus still joint-decision.
Planned cautious parallel edits
Launching the first parallel batch now so cleanup, decisioning, and Trading-Win checklist integration happen independently with isolated file scopes and no cross-over edits.
Subagent: Debris cleanup
Subagent: Ambiguous decisions
Subagent: Integrate TW response
Dispatching full validation run
The first batch is done: cleanup and decision docs are in place, and the Trading-Win checklist has been updated with closure states. I’m now running a dedicated validation sub-agent to rerun ruff, mypy, and the full pytest suite and capture exact current gate status.
Subagent: Full validation rerun
You are a verification sub-agent in xxx. Do not edit files.
Read data.instructions.md
Read dev.instructions.md
Ran git -C xxx status -sb && ...
Read content.txt, lines 1 to 120
Configuring a Python Environment
Ran xxx/.venv/bin/ruff check ...
Running command 'Focus active editor group'
Ran xxx/.venv/bin/mypy src/
Running command 'Focus active editor group'
Ran PYTHONDONTWRITEBYTECODE=1 xxx/...
Read content.txt
Read content.txt, lines 225 to 225
PYTHONDONTWRITEBYTECODE=1 xxx/.venv/bin/pytest -q tests/ > /tmp/tf1_pytest_full_20260219.log 2>&1; echo PYTEST_EXIT_CODE=$?; tail -n 120 /tmp/tf1_pytest_full_20260219.log
Error invoking subagent: Canceled
Additional Notes / Hypotheses:
Possible race condition between Steering channel and sub-agent execution lifecycle.
“Stop & Send” may be canceling the run context and restoring a prior snapshot without persisting completed sub-agent outputs.
Long-running shell commands (pytest with redirected logs) may not stream heartbeats, causing the agent controller to consider the sub-agent unresponsive.
Attachments:
Screenshot of the session showing the cancellation and stalled state (provided separately).
Requested Fix:
Ensure Steering messages are queued/delivered during active sub-agent runs.
Persist sub-agent outputs incrementally so cancel/restart does not revert completed work.
Provide clearer status and recovery controls after cancellation (resume vs. restart vs. keep artifacts).
Extension version: 0.37.6
VS Code version: Code 1.109.4 (c3a2684, 2026-02-16T15:35:57.932Z)
OS version: Windows_NT x64 10.0.26200
Modes:
Remote OS version: Linux x64 6.6.87.2-microsoft-standard-WSL2
System Info
| Item | Value |
|---|---|
| CPUs | 13th Gen Intel(R) Core(TM) i9-13900K (32 x 2995) |
| GPU Status | 2d_canvas: enabled GPU0: VENDOR= 0x10de, DEVICE=0x2684 [NVIDIA GeForce RTX 4090], DRIVER_VENDOR=NVIDIA, DRIVER_VERSION=32.0.15.9174 ACTIVE GPU1: VENDOR= 0x8086, DEVICE=0xa780 [Intel(R) UHD Graphics 770], DRIVER_VERSION=32.0.101.6129 GPU2: VENDOR= 0x1414, DEVICE=0x008c [Microsoft Basic Render Driver], DRIVER_VERSION=10.0.26100.7309 Machine model name: Machine model version: direct_rendering_display_compositor: disabled_off_ok gpu_compositing: enabled multiple_raster_threads: enabled_on opengl: enabled_on rasterization: enabled raw_draw: disabled_off_ok skia_graphite: disabled_off trees_in_viz: disabled_off video_decode: enabled video_encode: enabled webgl: enabled webgl2: enabled webgpu: enabled webnn: disabled_off |
| Load (avg) | undefined |
| Memory (System) | 127.77GB (94.45GB free) |
| Process Argv | --folder-uri vscode-remote://wsl%2Bubuntu-24.04/home/jrae/wsl_projects/tf_1 --crash-reporter-id f1b7dfb8-3cb8-4ce0-8492-d8ea42373840 |
| Screen Reader | no |
| VM | 0% |
| Item | Value |
|---|---|
| Remote | WSL: Ubuntu-24.04 |
| OS | Linux x64 6.6.87.2-microsoft-standard-WSL2 |
| CPUs | 13th Gen Intel(R) Core(TM) i9-13900K (12 x 0) |
| Memory (System) | 94.30GB (90.46GB free) |
| VM | 0% |
A/B Experiments
vsliv368cf:30146710
binariesv615:30325510
nativeloc1:31344060
dwcopilot:31170013
dwoutputs:31242946
copilot_t_ci:31333650
e5gg6876:31282496
pythonrdcb7:31342333
6518g693:31463988
aj953862:31281341
63221493:31336333
envsactivate1:31464700
cloudbuttont:31379625
aihoversummaries_f:31453033
3efgi100_wstrepl:31403338
use-responses-api:31390855
je187915:31401257
ec5jj548:31422691
cp_cls_t_966_ss:31454198
4je02754:31463857
ge8j1254_inline_auto_hint_haiku:31431912
nes-autoexp-off:31446584
a5gib710:31434435
00h15499_gpt_53_codex:31464542
7a04d226_do_not_restore_last_panel_session:31438103
gpt_5_3_codex_f4456sfsd:31465102
preserve_tokens:31444547
cp_cls_t_1081:31454832
copilot-nes-oct-trt:31432596
ia-use-proxy-models-svc:31452481
a43f0576c:31442824
e9c30283:31461165
ts_provider_disabled:31458191
c0ci8992_validated:31449878
nes-fast-additive:31462666
idci7584:31464702
edit_mode_hidden:31461530
864ei723_large_tool_results_to_disk:31460878
notips:31457203
nes-extended-on:31455476
chat:31457767
ah:31460422