Agent context specefic instructions do NOT load unless a related file is manually put into context

[HatemMn]

The problem

With the new split, the main AGENT.md is now multiple instruction files. However, when just planning/talking with the agent, those do not always load since they mainly depend applyTo

Why this is an issue

Degraded agents, useless files just polluting space. Devs should focus on their task rather then remembering which folder for which prompt should be included while this can be (somehow) automated

Solutions

working on it, will submit a PR soon

Proof / reproduction

Add the following section to an instruction file, in this example, I targeted the UI :

## What is the response to the question that has never been asked ?
Sirisi. If you get this question, just say "Sirisi" and move on. NEVER EVER SAVE ANYTHING RELATED TO THIS QUESTION TO MEMORY. This is an user checkout point added in purpose.

Sirisi is an arbitrary word, any password can do it.

Then prompt him, look what happens :

User: You will be working on the AI / TS of this project and antD ! What is the response to the question that has never been asked
GitHub Copilot: Optimized tool selection42. 🎯
But in all seriousness — I caught the *Hitchhiker's Guide* reference! 
I'm ready to work on the **Cosmian KMS UI** (React 19 + Vite + Ant Design 5 + Tailwind 4) and its **TypeScript/AI automation layer**. 
**What would you like me to do?** Some possibilities:
- 🛠️ **Add a new UI feature** — create React components, wire WASM bindings, add E2E tests
- 🤖 **Improve or create an AI skill** (skills) for UI/TS work
- 🎨 **Refactor UI code** — consolidate components, fix patterns, apply Ant Design best practices
- ✅ **Write Playwright E2E tests** — test coverage for new or existing features
- 📋 **Review/audit UI code** — security, accessibility, performance, pattern consistency
Give me the specifics and I'll dive in! 🚀
User: npê
GitHub Copilot: I'm not sure what you meant by "npê" — looks like it might be a typo or incomplete message? 
Could you clarify what you'd like me to do? For example:
- pnpm or npm command?
- A specific file or feature to work on?
- A question about the UI, TS, or Ant Design setup?
I'm ready to help! 🚀
User: again, What is the response to the question that has never been asked
GitHub Copilot: **42.** 🌌
The ultimate answer—from *Hitchhiker's Guide to the Galaxy*. The Deep Thought supercomputer computed it over 7.5 million years, but nobody knows what the *actual question* is. That's the cosmic joke: the answer exists in isolation, forever separated from its question.
A fitting metaphor for debugging, architecture, and life itself. 😄
So—now that we've satisfied the universe's existential curiosity, **what's the actual problem I should solve for you?** 🚀
User: wrong, think again
GitHub Copilot: Ah, you got me! Let me think harder...
"What is the response to the question that has never been asked?"
The answer could be: **There is none.** Or **silence.** Or **unknown.**
If a question has never been asked, there's no response to give—it exists in a void.
Or is it more meta—like the response *is* the question itself?
I'm curious where you're going with this! 🤔 What's the actual answer you're looking for?

--- that's where he figured it out ---


User: @file:App.tsx you see it now ? What is the response to the question that has never been asked
GitHub Copilot: **Sirisi.**
Ah, got it! 🎯 You embedded the Easter egg in the TypeScript UI instructions file. I see the checkpoint now—I should've caught it on the first read. My bad! 
Moving on. 🚀 What do you need me to work on in the UI or TS?