VS Code's Agent Host chat keeps spinning

“Follow up to https://github.com/microsoft/vscode/pull/341700, which streams a detached background shell's output from the Copilot SDK's session.rpc.tasks.readOutput into its tool call's terminal. Review of that PR found three things, none of which block it. The reload and restart gaps are already tracked in https://github.com/microsoft/vscode/issues/341645. lastSnapshot grows without bound for background shells NonPtyShellTerminalStreams.appendChunk appends every chunk to stream.lastSnapshot (co”

“[unanswered, 36 views] I’m using GitHub Copilot Chat in VS Code. Chat worked for a few minutes, then suddenly starts failing with the error below and stops responding. Error: Pending response rejected since connection got disposed at Object.dispose (c:\Users\ \AppData\Local\Programs\Microsoft VS Code\6d62b1d5da\resources\app\node_modules.asar\vscode-jsonrpc\lib\common\connection.js:1169:27) at markDisconnected (file:///c:/Users/ /AppData/Local/Programs/Microsoft%20VS%20Code/6d62b1d5da/”

“Reported on VS Code 1.141 stable and 1.143 Insiders with the bundled Copilot runtime. Steps to reproduce 1. Install a file based plugin through chat.pluginLocations containing a parent agent and a sibling agent enabled for subagent use. 2. Start a Copilot / Agent Host session using the bundled runtime and select the plugin's parent agent. 3. Ask the parent to delegate a task to the sibling agent. Expected: Eligible agents from the plugin remain available to the task tool alongside built in and w”

“Preflight Checklist [x] I have searched existing issues and this hasn't been reported yet [x] This is a single bug report (please file separate reports for different bugs) [x] I am using the latest version of Claude Code What's Wrong? What's wrong The automatic classification of conversations in agent view needs some work. I continuously get conversations stuck in the "Ready for review" section, and the only way to get them out of the way is to delete them. For example, I currently have three co”

“Copilot Chat Extension Version: bundled (VS Code 1.141.0) VS Code Version: 1.141.0 OS Version: Windows 11 (also present on a second machine with the same version) Feature: agent sessions running in the local agent host process (the mode picker option "Copilot (runs the Copilot SDK agent in a local agent host process)") Selected model: any BYOK model via a custom OpenAI compatible endpoint (verified through a local logging proxy) Logs: outgoing request timings captured at a local OpenAI compatibl”

“Copilot Chat Extension Version: Built into VS Code Insiders 1.142.0 insider VS Code Version: 1.142.0 insider OS Version: macOS arm64 Feature (e.g. agent/edit/ask mode): Agent Host custom agent picker and harness switching Selected model (e.g. GPT 4.1, Claude 3.7 Sonnet): Auto Logs: Agent Host debug export showed the affected SDK session using the generic base identity instead of the selected custom agent identity. Steps to Reproduce: 1. Open a new Copilot Agent Host chat and select a custom agen”

“Copilot Chat Extension Version: Not captured for this report. Agent Host SDK: 1.0.92 4.unstable.r37397886721.gad270aa . VS Code Version: 1.141.0 insider , commit 2a59476c9bfcb90b3ddc372c36762471b7dfad1c . OS Version: Windows. Exact OS build not captured for this report. Feature (e.g. agent/edit/ask mode): Copilot Agent Host, first turn of a new session. Selected model (e.g. GPT 4.1, Claude 3.7 Sonnet): Auto, resolved to gpt 6 luna . Logs: Sanitized timing findings below from an Agent Host debug”

“Does this issue occur when all extensions are disabled?: Not tested. VS Code Version: 1.141.0 insider on the host, commit ae1080a9ed1098a8b532c6cfd50977d784b29e6d , built October 5, 2026. The client was also running the latest October 5 desktop Insiders. OS Version: macOS 27.0 on the host; client OS not captured. Steps to Reproduce: 1. Make an existing agent host available through a remote tunnel and connect to it from desktop Insiders on another device. 2. Open a long running Copilot CLI sessio”

“Does this issue occur when all extensions are disabled?: Yes VS Code Version: 1.140.0 and later. It starts with 337621 and is not present in 1.139. OS Version: All Steps to Reproduce: 1. In an Agent Host Copilot session, select the HydraFusion model and send a request that runs a multi phase Fusion workflow. 2. With telemetry output logging on, look at the agentHost.turnCompleted events. 3. Each Fusion phase that gets a child chat produces a turnCompleted event with: isSubagentSession: true pare”

“Copilot Chat Extension Version: Not collected; affected sessions use the Agent Host Copilot provider ( copilotcli ). VS Code Version: 1.140.0 insider, commit b6c7324cedb6fc4ed639e3c7e1d4b233a09df8bc , build date 2026 09 29T20:21:41. OS Version: Windows, 10.0.26200. Feature (e.g. agent/edit/ask mode): Agents Window, local Agent Host session discovery/restoration. Selected model (e.g. GPT 4.1, Claude 3.7 Sonnet): Not applicable; these sessions have no persisted conversation turns. Logs: Relevant e”

“Copilot Chat Extension Version: 0.68.2026092904 VS Code Version: 1.140.0 insider (commit f9bb522f5746a56791aa4407a3db5c1b5cd3ad04) OS Version: macOS Feature (e.g. agent/edit/ask mode): Enterprise managed Local Copilot OTel policy recovery Selected model (e.g. GPT 4.1, Claude 3.7 Sonnet): N/A Logs: Before the repro, Local OTel was healthy: [OTel] Instrumentation enabled — exporter=otlp http endpoint=http://localhost:4318/ captureContent=false . The failed refresh settled the account policy gate a”

“Summary While investigating a sustained VS Code Insiders renderer CPU spike in a long Agent Host chat session, the Agent Host response file changes provider was identified as the root trigger: it re subscribes every historical request's file change diff observable to the whole chat's live streaming state, so every streamed token of the active turn re parses every prior turn's response into diffs again . This is a distinct, upstream root cause from 337244 (which covers the downstream Prompt Timel”

“Summary A user reports that a long running Agent Host Copilot session, used for over a week, no longer opens. Selecting it leaves the chat loading indefinitely. The user reports that newer and older smaller sessions still load. Two privately supplied diagnostic captures were analyzed. Both show successful SDK session resume followed by a non completing history retrieval. No user logs, conversation contents, local paths, or session identifiers are included in this issue. Affected versions Origina”

“Type: Bug A customer reports that this regression is affecting many developers after updating to the latest Stable release. VS Code version: 1.138.0 Surface: Local GitHub Copilot Chat harness (not Agent Host / Copilot SDK) Regression: Yes; reported after updating to the latest Stable release Steps to Reproduce 1. Open local GitHub Copilot Chat. 2. Select Agent mode. 3. Restart VS Code. 4. Reopen Chat. Actual The selected mode has reset to Ask . The user must manually select Agent again after res”

“Copilot Chat Extension Version: N/A (built in Agent Host) VS Code Version: 1.139.0 insider ( 046944034292b5479b4e9a50ad1a508033ffb64f ) OS Version: macOS arm64 Feature (e.g. agent/edit/ask mode): Agent Host, follow up in an existing Copilot session Selected model (e.g. GPT 4.1, Claude 3.7 Sonnet): GPT 5.6 Sol Copilot runtime: 1.0.84 5 Logs: Agent Host debug bundle captured locally; sanitized causal timeline is included below. Steps to Reproduce: 1. Open an existing Agent Host session with prior”

“Type: <b Bug</b Don't know how to reproduce — it happens at random intervals but won't stop happening. Discussed on reddit by some other users: https://www.reddit.com/r/vscode/s/PLxSRt6af2 I've tried reinstalling vscode (being sure to delete /.vscode and library files), reinstalling extensions, etc. VS Code version: Code 1.137.0 (645f29cc3176500b4b5762ba887cf2a7f0ffdf2c, 2026 09 08T13:43:59 07:00) OS version: Darwin arm64 25.6.0 Modes: <details <summary System Info</summary |Item|Value| | | | |C”

“Problem In Copilot Agent Host chat, the response finishes but the chat progress indicator keeps spinning while an attached asynchronous shell command remains alive. Later messages can be treated as steering messages even though the previous answer has finished. A long running preview server triggered this in a real conversation. Even a subsequent tool free response remained open for more than 48 minutes. Stopping the server restored normal turn completion. Environment VS Code Insiders 1.138.0, c”

“Before Submitting [x] I searched open and closed issues and discussions for an existing request. [x] I checked whether this already exists on the dev branch or latest source. [x] I understand that maintainers want a well written issue or discussion before any code pull request. [x] This request is not a security vulnerability. Problem I am using a token usage display plugin. It make a line with all of the metrics. When I use multiple LLMs, the width is too narrow to display the entire line and i”

“Does this issue occur when all extensions are disabled?: Yes / N/A for the web client. This reproduces in vscode.dev/agents in both Firefox and Chromium, so it does not appear browser specific or extension specific. VS Code Version: 1.136.1 ( a44adf7f53e00964ab890f9f8758a334f1fc15bc ) OS Version: Linux x64 remote tunnel host Client: vscode.dev/agents Connection: VS Code tunnel Reproduced in: Firefox and Chromium Summary Opening one specific, relatively large Agent Host conversation in vscode.dev”

“Type: Bug Regular VS Code Chat Editor using the Copilot Agent Host harness. Discussion and screenshots: https://vscodeteam.slack.com/archives/C0AJZ5319PZ/p1788381749454329 Reproduction 1. Open the VS Code repository in an Insiders editor window. 2. Open a fresh Chat Editor using the Copilot harness. 3. Select the real Data custom agent. 4. Send hi . This reproduces again for every fresh session. After the first turn has completed, selecting Data again sticks for subsequent turns. Actual The firs”

“Description Summary The Copilot CLI chat sessions provider ( CopilotCLIChatSessionContentProvider ) re enumerates the entire Copilot CLI session store on every refresh — no incremental update, cache, or cap. provideChatSessionItems (logged as listSessions took <n ms ) calls getAllSessions() and maps every session. It's triggered constantly: a 5s poll runs while any CLI session is active ( setInterval(updateActiveSessionsOnly, 5000) ), and any session state change fires onDidChangeChatSessionItem”

“Building on the recent work to have chat.tools.terminal.autoApprove and chat.tools.terminal.enableAutoApprove respected for Agent Host Copilot ( 322633, 322774), I'd like to bring back the more granular, "sticky" approval controls we used to have so users aren't re prompted for commands they've already OK'd. Issue: Today the AH terminal auto approver ( src/vs/platform/agentHost/node/commandAutoApprover.ts ) only compiles the forwarded chat.tools.terminal.autoApprove rules and checks each command”

“Copilot Chat Extension Version: 0.42.3 VS Code Version: 1.114.0 OS Version: Feature (e.g. agent/edit/ask mode): agent Selected model (e.g. GPT 4.1, Claude 3.7 Sonnet): GPT 5.4 Logs: Steps to Reproduce: 1. give instructions to copilot that result in a TODO list 2. observe how the 1st one shows as in progress and nothing else happens even though copilot works through them...”

“When having copilot and Modus toolbox extensions both installed, at times when using the copilot chat extension for a while, when reloading Window and starting the Modus Toolbox extension, it will hang when trying to do node fetches due to copilot hogging all resources Copilot Chat Extension Version: V0.38.2 ModusToolbox Assistant V2.27.0 VS Code Version: 1.120.0 <img width="367" height="292" alt="Image" src="https://github.com/user attachments/assets/4dc91349 4907 49eb b2eb 5baea44f7e43" / OS V”

“Testing 301037 After enabling github.copilot.chat.agentDebugLog.enabled , the /troubleshoot command appears as expected. Subsequent disabling of the setting & reloading of VS Code still renders the /troubleshoot command in chat i.e. in the blue treatment. Executing the command prompts me to enable the required settings, as expected. I would expect the agent to forget about the skill if the settings were disabled.”

“Copilot Chat Extension Version: 0.37.6 VS Code Version: 1.109.4 OS Version: Windows 11 (+devcontainer) Feature (e.g. agent/edit/ask mode): Plan mode Selected model (e.g. GPT 4.1, Claude 3.7 Sonnet): n/a Steps to Reproduce: 1. Create a plan using plan mode 2. Observe that section "proceed from plan" is visible. User can click on "start implementation" 3. Navigate back to chat overview 4. Click on plan session 5. Proceed from plan section is not visible even though plan mode is still selected Work”

“Copilot Chat Extension Version: 0.33.2 (and GitHub Copilot is 1.388.0) VS Code Version: Version:1.106.2 (user setup) OS Version: Windows 11 Pro (25H2 | 26200.7171 | Windows Feature Experience Pack 1000.26100.265.0) Feature (Ask(and then within Geminis longer thinking/reasoning)): Selected model: Gemini 3 Pro(Preview) Logs: Chat Debug Output: copilot all prompts 2025 11 25T00 47 03.chatreplay.json (Line 1546 starts prompt) Output Trace Level(prompt timestamp start@ line 14577): output gemini loop”

“Copilot Chat Extension Version: 0.29.1 VS Code Version: 1.102.3 OS Version: MacOS Sequoia 15.6 Feature (e.g. agent/edit/ask mode): Agent Mode Selected model (e.g. GPT 4.1, Claude 3.7 Sonnet): Gemini 2.5 Pro Logs: I used Gemini 2.5 Pro and it got stuck in an infinite loop. <img width="370" height="733" alt="Image" src="https://github.com/user attachments/assets/4bef882a a2e2 497e b9de 8e86d56adf52" /”

“Copilot Chat Extension Version: 1.293.1471 VS Code Version: Version: 1.99.0 insider (user setup) Steps to Reproduce: 1. Agent Mode, Claude 3.5 2. Request multiple actions be completed 3. Some Actions will be completed, but after a random period of time, will return error "Sorry, no response was returned." Upgraded extension version in case the issue was with the version i was on from last night. No difference.”