IDEA intellij - copilot cannot see the output in terminal of command it executed

[dantuch]

Select Topic Area

Bug

Copilot Feature Area

Copilot Agent Mode

Body

I made copilot try a basic:
echo "This is a test output to see if I can view it"
and it didnt see the output in my terminal - is that normal?

"I see the issue now. It appears that while I'm executing commands in the terminal, I'm not able to see their actual output content. This applies not just to Spectral but to basic commands like echo as well. The function_results are showing me that commands were executed, but the actual output text isn't being displayed within the Output: section.
This explains why I've been having trouble verifying the Spectral validation results - I can confirm that commands are running but I can't see their actual textual output in the terminal.
Given this limitation, let's modify our approach. Instead of relying on direct terminal output visibility, we can use the read_file function to check the contents of the validation output "

copilot: 1.5.45.10445-243-nightly
IDEA: Build #IU-251.25410.129, built on May 9, 2025
tested on Mac: 15.0.1 (24A348)

btw. my copilot in MOST cases doesnt see the terminal output - but sometimes it does see it - so it seems not deterministic to me.

[Akash1134]

Hi @dantuch,

Thank you for reporting this issue with GitHub Copilot in agent mode not being able to reliably see terminal output in IntelliJ IDEA. This is a known limitation in the current implementation of Copilot's agent mode for JetBrains IDEs.

Issue Details

From your report, you're experiencing:

• Copilot agent mode cannot consistently see the output of terminal commands it runs
• You tested with a basic command (echo "This is a test output to see if I can view it") with the same result
• The behavior appears to be intermittent - sometimes it can see the output, but most times it cannot

Technical Explanation

This behavior occurs due to how Copilot agent mode interfaces with the JetBrains terminal. When Copilot executes a command in the terminal:

1. It can confirm the command was executed (which explains the "function_results" confirmation you mentioned)
2. However, the mechanism for capturing the actual terminal output text in JetBrains IDEs is not consistently reliable

The intermittent nature of the issue (sometimes working, sometimes not) is likely related to timing and buffering differences depending on the specific terminal session, command type, and output volume.

Current Workarounds

Until this is addressed in a future update, here are some effective workarounds:

1. File-based Output Redirection:

   • Modify commands to redirect output to files that Copilot can then read
   • Example: echo "Test output" > output.txt && cat output.txt

2. Alternative Command Verification Methods:

   • For validation operations (like your mentioned Spectral validation), have Copilot check for exit codes instead of trying to read output text
   • Use commands that create artifacts Copilot can verify independently

3. Manual Output Checking:

   • For critical terminal operations, you may need to check the output yourself and provide the results back to Copilot in your next prompt

Thank you for your patience and for helping us improve GitHub Copilot!

[rnorth]

@Akash1134 thanks for these workarounds, but they're pretty intrusive and aren't practical to do for every single agent operation.

I notice it's now a few months since this discussion was raised and it's still a problem. Please could you provide some clearer updates on when this will be tackled?

[Dave-Vaughan]

Is still the case after updating this morning. vscode can do this just fine - infact sometimes its scary how close to a professional hacker the speed of it is...., but i really dont want to use vscode as i'm used to the tools in intellij.

[adrielisa]

wow

[mikkelstaunsholm]

it also randomly forgets some characters (the pipe is there in the file):

There is a difference on line 24. The file on disk has:
location ~* .(jscss) {

But the attached version has:
location ~* .(js|css) {

Quite frankly bizarre that this is not fixed yet... Is it a plot to move people to vscode :-) ?

[github-actions[bot]]

💬 Your Product Feedback Has Been Submitted 🎉

Thank you for taking the time to share your insights with us! Your feedback is invaluable as we build a better GitHub experience for all our users.

Here's what you can expect moving forward ⏩

• Your input will be carefully reviewed and cataloged by members of our product teams. 
  • Due to the high volume of submissions, we may not always be able to provide individual responses.
  • Rest assured, your feedback will help chart our course for product improvements.
• Other users may engage with your post, sharing their own perspectives or experiences. 
• GitHub staff may reach out for further clarification or insight. 
  • We may 'Answer' your discussion if there is a current solution, workaround, or roadmap/changelog post related to the feedback.

Where to look to see what's shipping 👀

• Read the Changelog for real-time updates on the latest GitHub features, enhancements, and calls for feedback.
• Explore our Product Roadmap, which details upcoming major releases and initiatives.

What you can do in the meantime 💻

• Upvote and comment on other user feedback Discussions that resonate with you.
• Add more information at any point! Useful details include: use cases, relevant labels, desired outcomes, and any accompanying screenshots.

As a member of the GitHub community, your participation is essential. While we can't promise that every suggestion will be implemented, we want to emphasize that your feedback is instrumental in guiding our decisions and priorities.

Thank you once again for your contribution to making GitHub even better! We're grateful for your ongoing support and collaboration in shaping the future of our platform. ⭐

[amoerie]

I've had this issue today on Windows. Changing the terminal engine to "Classic" and the shell to "CMD.exe" fixed things for me.

[rpalacios]

I have same problem too.

I tried it with engine "Classic" and powershell 7, and it worked fine.

[adrielisa]

interesting

[Arham4]

In VS Code (in the context of Java) and running tests I generally tell copilot to "use test runner for java to run tests." Then it can read the results of tests. Wonder what the alternative is for IntelliJ.

[hmrguez]

Does anyone have a workaround for macos? Not very keen on using powershell for the time being

[davidcarma]

Solution: Fix Copilot Agent's Terminal Output Issue on macOS
Problem:
When using GitHub Copilot Agent in Visual Studio Code on macOS, the agent may fail to read terminal command output, causing it to hang or provide incorrect responses. This is often related to the default shell being zsh.

Solution:
The issue can be resolved by changing the default terminal shell in VS Code from zsh to bash. This provides an environment that the Copilot Agent can reliably interact with.

Step-by-Step Instructions
Open the Command Palette in VS Code by pressing Cmd+Shift+P on your keyboard.

In the search bar, type Terminal: Select Default Profile.

From the list of available profiles, select "bash".

Close and reopen any active integrated terminals to apply the change.

After following these steps, the next time you use the Copilot Agent, it will use the bash shell, which should allow it to correctly read command output.

Why This Works:
While both bash and zsh are powerful shells, they have subtle differences in how they handle command output, scripting, and I/O redirection. The Copilot Agent's underlying logic may have a more robust or specific integration with bash that is not fully compatible with zsh, leading to the observed behavior. Switching to bash provides a known working environment for the agent.

[HWiese1980]

It seems in IntelliJ IDEA I can trigger this bug by actively using the Copilot terminal session for manual commands myself. I can "fix" the issue by closing the session and let Agent open a "clean" new one.

[brolland-byg]

Having the problem with bash under wsl-linux-Ubuntu-bash
it is getting annoying because starting a new session costs a new premium request and loose the history.
Is there any way to step back to an older plugin version ?

[yuv-c-netapp]

Sorry @Akash1134 and Intellij - but the suggestions (pressing enter after a cmd, or changing theme to zsh) are either degrading the capabilities in agent mode or restrict the dev from using its own environment.
For example, podman isn't accessible in my bash shell, but is from zsh.
So this needs to be addressed.
Please see the answers in this issue (vs code). I am personally not using p10k, so I don't think its relevant for me, but its a starting point.

Please fix this. This is a paid product and should support working with the most common shells and coding assitants.

[lmhajdukd]

Same issue, Copilot doesn't see npm test results - failing tests and stacktraces after asking it to run them. I'm not using p10k, pure bash. Still doesn't work. Paying for a license for no reason.

[ChrisDugdale]

Same for the the command runs, then just hangs waiting until I cancel it. This is in Rider using copilot agent mode.

[alankpax]

Has same issue with Phpstorm on Windows with default ubuntu WSL 2 bash.

[kpihlblad]

Same for the the command runs, then just hangs waiting until I cancel it. This is in Rider using copilot agent mode.

It has been like this consistently since akways. I had to disable the terminal tool or every session inevitably ends up in a unresponsive terminal. Exiting bash or closing the terminal doesn't help.

Something also occasionally dumps this

Error parsing input: java.lang.IllegalStateException: java.io.IOException: Error reading ps output:your 131072x1 screen size is bogus. expect trouble

I'm using bash and and are running in WSL

[jimpudar]

I am still seeing this issue on Rider 2025.2.2 running on Arch Linux.

Sometimes it seems to be able to see the command output, but very often Copilot says:

The command seems to be running but not returning output.

I'm using the "Reworked 2025" terminal within Rider.

[jimpudar]

As others have suggested, if I close the copilot terminal and never interact with it, this problem seems to not occur.

[CaptainKeyboard]

Still happening in IntelliJ on the latest version. Pretty annoying when it builds a gradle project. When it sees nothing, it assumes the build was successful, while it often isn't.

[andi-dev]

Still running into this issue in IntelliJ on Mac as well. This is effectively degrading the performance enormously, and I seriously consider to switch back to Junie.

[andi-dev]

Possibly as someone else mentioned here before, it is working if I entirely avoid to interact with the copilot-terminal at all.

[Renevdo]

Still a problem on the latest version on Windows

[rbnzdave]

Just chceked on this in 2025.3.1 and getting copilot to run a git status and a make a directory called dave and give everyone read access and both worked exactly as hoped, this was not the case previously

While that worked, I asked it to get a current directory listing, and it tried getting a path ( am running the project in a wsl install of ubuntu 22.04 so is linux pathing with a windows install of intellij) and it failed 3 times with the directory tool, tho it did try with 3 variants of the prompts

But i need to add that when i got it to do a directory listing, without mentioning the project, it asked to run ls -l and then when i asked again it told me i already had the listing. I think its finally fixed.

[rbnzdave]

side note the project isnt in C: anything, so a bit misguided there.

[crash83k]

I find that clearing the terminal buffer (or closing it and forcing copilot to open a new one) greatly mitigates this issue.

The issue persists to this day. (Pop!_OS 22.04)

[mik3fly]

The issue is still present - february 9 2026 / Intellij IDEA 2025.1.3

[taewonee]

hey pals,
according to all of your comments, basically it's related to terminal settings.

just try to switch your terminal shell path

if bash works, then try to re-install or do something for your zsh
if zsh works, then why not use zsh?

[alankpax]

Well you shouldn't be forced to use some shell or terminal yourself to be able to use an AI assistant. And if this is related to settings what would be these settings ?

It works much better with VS Code plugin version. The quality of integration within Jetbrains IDEs of this feature make me consider dropping them. As much as I like them this is how it is now as so much things are going fast (skills, sub agents, coordination...).

[taewonee]

totally Agreed. I am considering Claude Code. just waiting for corporate's permission(and support) haha

well, since I was struggling with IDE problem, just sharing my experience.
good news, GPT restored all zsh settings (by back-up & re-install) in just 20 mins, and now zsh works with intelliJ+copilot.
must be helpful for somebody for sure

[alankpax]

Well I'm also considering Claude Code but my organization first chose Copilot so maybe I would not have the choice.
Using a TUI client as Claude Code may be an alternative to keep using Jetbrains IDE while using Claude Code from the terminal but that would be quite disapointing not having a complete integrated experience into the IDE.

note : the problems also happen with Powershell

[HWiese1980]

hey pals, according to all of your comments, basically it's related to terminal settings.

just try to switch your terminal shell path

if bash works, then try to re-install or do something for your zsh if zsh works, then why not use zsh?

It is not. I am using bash and still have the same problems

[mik3fly]

same, not related to terminal. It can works at beginning, but stop working after a while. i use my terminal hundred of times per day, and i don't have any problems. I suspect problems with copilot. Le jeu. 12 févr. 2026 à 16:38, Hendrik Wiese ***@***.***> a écrit :

…

hey pals, according to all of your comments, basically it's related to terminal settings. just try to switch your terminal shell path [image: image] <https://private-user-images.githubusercontent.com/56881580/548743093-a272f31b-b755-4c12-af8b-389b1e2c101c.png?jwt=eyJ0eXAiOiJKV1QiLCJhbGciOiJIUzI1NiJ9.eyJpc3MiOiJnaXRodWIuY29tIiwiYXVkIjoicmF3LmdpdGh1YnVzZXJjb250ZW50LmNvbSIsImtleSI6ImtleTUiLCJleHAiOjE3NzA5MTA4NTUsIm5iZiI6MTc3MDkxMDU1NSwicGF0aCI6Ii81Njg4MTU4MC81NDg3NDMwOTMtYTI3MmYzMWItYjc1NS00YzEyLWFmOGItMzg5YjFlMmMxMDFjLnBuZz9YLUFtei1BbGdvcml0aG09QVdTNC1ITUFDLVNIQTI1NiZYLUFtei1DcmVkZW50aWFsPUFLSUFWQ09EWUxTQTUzUFFLNFpBJTJGMjAyNjAyMTIlMkZ1cy1lYXN0LTElMkZzMyUyRmF3czRfcmVxdWVzdCZYLUFtei1EYXRlPTIwMjYwMjEyVDE1MzU1NVomWC1BbXotRXhwaXJlcz0zMDAmWC1BbXotU2lnbmF0dXJlPWFjYWJiNWRiNTMwN2E5ZDU4ODE0NjQ3MTBkYjcwMTE3YzFiNjMxMmMxNzI1MWU5YjNjNTVkMzkxYzljY2MwYjcmWC1BbXotU2lnbmVkSGVhZGVycz1ob3N0In0.w4ffewOoIs4O_kVQvzQjziyBfyMIY-vl6MfHp_tBb8c> if bash works, then try to re-install or do something for your zsh if zsh works, then why not use zsh? It is not. I am using bash and still have the same problems — Reply to this email directly, view it on GitHub <#160060 (reply in thread)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AQUPBF5JJG5RMBGAAP2A6FL4LSNANAVCNFSM6AAAAAB5X4QL5WVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTKNZYGQ4DSMQ> . You are receiving this because you commented.Message ID: ***@***.***>

-- MARTIN Bryan ***@***.*** 06 66 26 88 20 (Rejet des appels masqués)

[rekinariumado]

Confirming the issue is still present.
Spent multiple hours and a lot of unnecessary premium calls, because of this issue.
Originally I thought that it might be related to extra shell plugins (sdkman, omz, etc.), but even after disabling them all, I have intermittent "terminal desync", where the LLM (as they claim in the conversation, when asked about it) receive no output at all.
Tried many of suggested workarounds ("| cat", writing and reading a temp file, zsh/bash switch, closing the copilot terminal tab) - nothing works reliably in longer conversations.
By longer I mean literally asking to fix a single test and runing "./gradlew integrationTest" the first time; so in practice it's breaking on literally first error check attempt.
As a workaround, I currently instruct models to use specific redirection "> output-log.txt" that I have opened up in intellij's editor directly. Its seems that the output can be correctly passed this way. I'm attaching a screenshot of a sample session, where the model recognised that.

[epic-64]

You are dropping the ball. As of Feb 2026 the latest version of intellij ultimate still has this issue. 90% of the time copilot cannot read from the terminal. My coworkers who use VSCode are making fun of me and I'm legitimately starting to feel ashamed for having a Jetbrains subscription.
I have a bog standard ubuntu 24.04 setup with a bog standard bash shell. Been having this issue for almost a year now (since ai agent mode existed) and it's truly mind-boggling how bad the quality is.

[crash83k]

Each time the AI uses the console, clear the buffer when it finishes. It sucks, but it definitely works. The problem appears to be related to the amount of history still in the buffer.

But you're right. For a paid product, it's not OK for this to be an issue that still exists for this long.

[ky0n]

how do i clear the buffer?

[crash83k]

Easiest way is to right-click in the terminal and click "clear buffer". I haven't tested a prompt telling the AI to clear the terminal buffer ahead of time yet, but I bet that could help if it knows how.

[crash83k]

BTW, I just tested that it can clear the buffer after each CLI usage by telling the prompt "Clear the terminal buffer before each CLI/terminal use."

So far it's working.

[OscarVanL]

Thanks for the tip. This seemed to fix it for me. Claude is now clever enough to then do printf '\033c' && $command, which clears the terminal (TIL about this ANSI escape sequence used in terminal emulators to "reset to initial state").

Now the agent can read the outputs which it was struggling to do previously.

Hopefully someone from Microsoft addresses the root cause so we don't need to do this.

[ErikErkelens]

Submitted logs to JetBrains support, they beleive the issue is in the plugin. I've tried the workarounds listed above, and then some. The only one that works for me is to use both VSCode (for AI) and Rider (to manually edit and debug) - painful.

From what I can see, the dotnet test / build commands are actually being started correctly in Rider’s terminal. The IDE itself is not hanging, and the terminal process is running normally.
However, right after Copilot triggers these commands, the IDE logs show terminal buffer access errors, and GitHub Copilot’s agent appears to receive an empty result from the terminal.
This matches a known issue on the Copilot side, where the agent can run commands but intermittently fails to read the terminal output in JetBrains IDEs. When that happens, Copilot assumes the command produced no output and retries with variations like the ones you’re seeing.

So based on the evidence, Rider is executing the commands correctly. The issue is in GitHub Copilot’s terminal integration/output capture, not in Rider’s test runner or compiler.
The GitHub Copilot team member posted a few workarounds you can try.

If the problem persists, I recommend reporting it directly to the GitHub Copilot team, as it appears to be on their side of the integration.

Kind regards,
<>

JetBrains
https://www.jetbrains.com/
The Drive to Develop

[johan313]

This is still an major issue and is wasting insane amount of tokens. Is there a bug ticket opened for this? Not the quality I would expect from a paid product with millions of users.

[rbnzdave]

Well here's a new one. The problem is gone for me, as I mentioned earlier. I also unsubscribed then. Good job that did. Even says down the bottom that I'm not subscribed and can subscribe. I am bombarded with this thread all the fn time. If its not one thing its another.

[Yoann-Abbes]

Hi
Here is a "golden rule" I'm using that seem to handle the problem and avoid waisting to much token over this issue when running Copilot on IntelliJ

Golden Rule — Terminal output in JetBrains (Agent Mode)

At the very start of every task and after every model change, run this self-test before doing anything else:
1. Run echo "TERMINAL_CHECK_OK" with isBackground: true
2. Call get_terminal_output on the returned ID
3. If the output contains TERMINAL_CHECK_OK → terminal is working, proceed normally with isBackground: true for all commands
4. If the output is empty or missing → terminal is broken in this session, stop and tell the user to clear the terminal buffer manually (right-click → "Clear Buffer" in JetBrains) then retry the self-test

Rules that always apply regardless:

- Always use isBackground: true — never isBackground: false — for every terminal command
- Always read output via get_terminal_output on the returned ID
- Never prefix commands with printf '\033c' — it wipes the command's own output
- isBackground: false returns empty output 100% of the time in JetBrains — it is permanently broken

Why this matters: This is a known JetBrains + GitHub Copilot bug. It affects all models. The self-test at task start ensures the issue is caught immediately rather than silently wasting prompts on empty terminal reads.

I'm not sure about the efficiency on long term, don't hesitate to give feedback if you test

[KIVagant]

A few steps that helped me to deal with the problem:

1. In the Settings > Tools > Terminal:

• Terminal Engine: keep the Reworked 2025 or try Classic if it does not help.
• Application Settings > Shell path: /bin/bash -l (not ZSH!)

2. I have translated ~/.zshrc settings into the ~/.bashrc so $PATH is equal in both shells. This is crucial.
3. ~/.bashrc cannot be automatically picked by IntellijIDEA (when the app itself is launched directly and not from a terminal) as it inherits environment from launchd. And launchd does not read ~/.bashrc, ~/.zprofile, etc. But, it reads ~/.bash_profile, so add these lines there:

# ~/.bash_profile
if [ -f ~/.bashrc ]; then
  . ~/.bashrc
fi

4. Log out and log in, it won't work otherwise.
5. Launch IntellijIdea and check that a new Terminal session starts with bash and $PATH and other crucial environment variables are available. Ask Copilot to validate access to cli output.

Additionally I recommend this:
Settings > Advanced settings > Terminal:

• Terminal scrollback buffer size: increase, i.e. 8000 lines.
• New terminal output capacity: increase, i.e. 4098 KB

If you still encounter the problem, a remediation is to run clear in the Copilot-initiated terminal window in between commands from time to time.