System prompt should not prevent creating markdown files in markdown-heavy repos
- Langage dominant
- Shell
- Étoiles
- 11.2k
- Forks
- 1.9k
- Merge moyen
- 14 h 16 min
- PR mergées (30 j)
- 6
Description
## Problem
The Copilot CLI agent's system prompt includes an instruction: *"Do not create markdown files in the repository for planning, notes, or tracking."*
This is too broad. When a user explicitly asks the agent to create markdown files in a repo that already uses markdown as its primary content format (docs, communication assets, design specs), the agent instead hides the files in an invisible session workspace. The user then can't find them and has to ask why.
## What happened
I asked the agent to create two .md files (an overview doc and a shipping plan) in a repo that already has 5+ markdown files in its `docs/` directory. The agent put them in `~/.copilot/session-state/` instead of the repo. When I asked why, it cited this system instruction.
## Suggested improvement
The instruction should be context-aware:
- If the repo already has markdown files as deliverables (not just a README), respect that pattern
- If the user explicitly asks to create a file, put it where they can find it (the repo) rather than a hidden session directory
- The current rule makes sense for avoiding litter in code-heavy repos, but it shouldn't override explicit user intent
## Environment
- Copilot CLI v1.0.41
- Running in a repo with `docs/` directory containing multiple .md files
Guide de contribution
Ouvrir le guide de contribution
Piste de recherche
Start by locating the Copilot CLI system prompt containing the markdown-file instruction and the handling of ~/.copilot/session-state/. Reproduce the request in a repository with multiple docs/*.md deliverables, then compare it with a code-focused repository. Done means explicit markdown-file requests are placed in the repository when its content pattern supports them without removing the existing anti-litter behavior.
Rédigé par le modèle d'indexation à partir du texte de l'issue.
Évaluation
- Stack technique
- shell
- Domaine
- ai, cli
- Type d'issue
- Fonctionnalité
- Difficulté
- 4/5
- Temps estimé
- 3-5 jours
- Activité
- Calme
- Clarté
- Plutôt claire
- Accessibilité débutants
- 52/100