Copilot no longer reading the instructions file?

[jannes-io]

🏷️ Discussion Type

Bug

💬 Feature/Topic Area

Copilot CLI

Body

2 weeks ago, usage of copilot CLI was a breeze, things were working well and the agents always respected everything noted in .github/copilot-instructions.md

But since last week, it's been going off the rails constantly and doing things that contradict instructions in the instructions file.

To give some examples:

• it suddenly "forgot" how to run tests, even though all the commands are in the instructions file, it worked flawlessly before
• we have an explicit "Avoid comments explaining what code does", and it has always respected that and kept comments to a bare minimum, but now it starts writing essays again in function/class docblocks

Still using the same models as before, so I believe the issue is somewhere in some update of the CLI harness?

[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. ⭐

[kamrullab]

I've noticed similar behavior recently.

It doesn't seem to be limited to your setup. There have been a few reports from other users about Copilot CLI becoming less consistent with custom instructions after recent updates, especially around copilot-instructions.md and instruction handling.

If you're still using the same models and your instruction file hasn't changed, it does sound like something may have changed in the CLI rather than in your repository.

Hopefully the team can take a look and confirm whether this is an unintended regression.

[initial-d]

This is worth reporting as a possible regression, but I would try to turn it into a very small reproduction because instruction handling varies a lot by Copilot surface.

The relevant docs say that Copilot CLI supports repository-wide custom instructions from .github/copilot-instructions.md, and that custom instructions are automatically included while working in that repository:

• https://docs.github.com/en/copilot/how-tos/copilot-cli/customize-copilot/add-custom-instructions
• https://docs.github.com/en/copilot/how-tos/copilot-cli/use-copilot-cli/overview#use-custom-instructions
• https://docs.github.com/en/copilot/reference/custom-instructions-support

A good bug report would include:

1. The Copilot CLI version before/after, if you can find it from shell history or package manager logs.

2. The exact working directory when you start the CLI. If the CLI is launched from a submodule, sibling worktree, symlinked path, or outside the repo root, instruction discovery may be the first thing to rule out.

3. A minimal instruction that is impossible to miss, for example:

   Always answer with the exact phrase INSTRUCTION_FILE_WAS_READ before doing anything else.

4. A minimal prompt that should trigger it, run from the repo root.

5. Whether the same repo works in VS Code / JetBrains Copilot Chat but fails only in CLI. That would separate a CLI harness regression from a general Copilot-instructions issue.

As a temporary workaround, I would move the most critical commands to the top of .github/copilot-instructions.md in very direct language, for example:

Before changing code, run tests with: npm test
Do not add explanatory comments unless the user explicitly asks for them.

If your Copilot surface supports path-specific or agent instructions, you can also duplicate critical rules into .github/instructions/*.instructions.md or an agent instruction file, but I would still keep the minimal repro focused on .github/copilot-instructions.md because that is the documented repository-wide CLI path.

[jannes-io]

If the Copilot CLI was an open-source project where we had the ability to inspect source, see actual changes and could locally reproduce issues, I'd totally agree.

But it's not. So the responsibility of listening to customer feedback, identifying bugs, and doing a proper root cause analysis is not on us, it's on the copilot team. Kind of odd that the biggest host of open-source software keeps everything closed-source...

Copilot CLI also seems to auto-update constantly, like a big part of it is just a TUI wrapper around a web application (I didn't have to update when they switched to AI credits, or to get newly released models). So I have no idea where to even look for update logs or similar, it includes some self-updater on launch.

The instructions worked well ~2 weeks ago. Now it keeps going off the rails. There's my bug report.. Fix it 🤷‍♂️

[pravendrarj]

This does sound like a possible instruction-loading regression, especially if the same repository, model and instruction file worked consistently before.

I would test it with a minimal repository and one unmistakable instruction, such as:

Before performing any task, respond with: INSTRUCTIONS_LOADED

I would also capture:

Copilot CLI version
Operating system and installation method
Working directory from which Copilot CLI was started
Location of .github/copilot-instructions.md
Model used
A transcript showing the instruction being ignored

It is also worth checking whether the CLI is being launched from the repository root rather than a parent directory, submodule, worktree or symlinked location.

GitHub documents that repository-wide instructions in .github/copilot-instructions.md should be automatically included when Copilot CLI works inside that repository. If the minimal reproduction fails consistently, I would report it to GitHub Support as a regression and attach the diagnostic information.

Documentation: https://docs.github.com/en/copilot/how-tos/copilot-cli/customize-copilot/add-custom-instructions