Hacktoberfest 2026 : les issues que les mainteneurs ont marquées pour octobre, ouvertes et accessibles aux débutants. Parcourir les issues Hacktoberfest

Selecting a foreground custom agent leaves the session model unchanged (Windows 1.1.23)

Ouverte
  • Quelque chose ne va pas ?
#4,097 0 commentaires 0 réactions 0 personnes assignées Voir sur GitHub

Personne n'a encore pris cette issue.

Évaluation

Difficulté
4/5
Temps estimé
3-5 jours
Accessibilité débutants
55/100
Type d'issue
Bug
Clarté
Plutôt claire
Activité
Active
Stack technique
github
Domaine
ai, desktop

Piste de recherche

Start by reproducing the issue in the GitHub Copilot desktop app with the repository's .github/agents/apex.agent.md profile, checking the model picker before and after selecting APEX. Compare the effective model with standalone Copilot CLI, then determine whether the mismatch is only in the picker or also affects inference; done means the declared model is applied or precedence differences are documented.

Rédigé par le modèle d'indexation à partir du texte de l'issue.

Description

Short summary

Selecting a repository custom agent in the desktop app leaves the session model unchanged, despite an explicit
model: in the agent profile. The same coordinator profile switches models in standalone Copilot CLI.

Affected version or release

GitHub Copilot app 1.1.23, verified through Windows installation metadata.

Installation context
  • Windows 11 x64, build 26200.
  • Local repository, Interactive session, repository agents under .github/agents/.
  • Comparison client: standalone Copilot CLI 1.0.86 running in Ubuntu/WSL2.
  • The desktop app's embedded engine version has not been established. This is not an identical-engine comparison.
What happened?

After selecting APEX, the app showed both APEX and GPT-6 Astra / Medium. Sending a short prompt started the
session and returned a response, but the picker still showed Astra. MAI-Code-1.1-Flash was available in the model picker.

Selecting APEX in standalone CLI changed the model to mai-code-1.1-flash. In the desktop app, manually selecting MAI
worked and allowed us to continue testing. We did not need to replace the agent's model setting with Auto.

The observation concerns the session's displayed selection. We have not captured inference telemetry proving whether
the mismatch is limited to the UI or also affects the actual model request.

Steps to reproduce

Our installed .github/agents/apex.agent.md begins with:

---
name: APEX
description: Fast coordinator for APEX status, resume, and direct specialist handoff.
target: github-copilot
model: mai-code-1.1-flash
user-invocable: true
disable-model-invocation: true
tools:
  - ask_user
  - apex/status
  - apex/nextTask
  - apex/projectCreate
  - apex/projectList
  - apex/projectUse
  - apex/projectDelete
  - apex/gateDecide
---

This is the tested profile excerpt, not a separately verified minimal reproduction; its body contains APEX workflow
instructions and its project has an APEX MCP server.

  1. Open the repository in the desktop app and start an existing local repository session in Interactive mode.
  2. With GPT-6 Astra selected, select APEX using /agent or the agent picker.
  3. Check the model picker before and after a short prompt such as test.
  4. Observe that APEX is selected but the model picker remains on Astra.
  5. In standalone CLI, select APEX with the same coordinator declaration and observe the switch to MAI.

We initially encountered a separate missing-agent error in a new worktree. The model-selection observation above
persisted after creating a working local-repository session; this report is not about that startup error.

Expected behavior

Selecting a custom agent should apply its declared model when available, or clearly explain that an explicit session
selection takes precedence. The picker should show the effective model. If desktop and CLI intentionally use different
precedence rules, please document those differences.

Additional context
Langage dominant
Aucune donnée de langage
Étoiles
2.2k
Forks
174
Aucune PR mergée en 30 j

Préparer son environnement

Par où commencer

  1. Lisez l'issue en entier, puis le guide de contribution du projet.
  2. Signalez en commentaire que vous la prenez — cela évite que deux personnes fassent le même travail.
  3. Forkez le dépôt et travaillez sur une branche.
  4. Ouvrez une pull request qui référence le numéro de l'issue.

Autres issues de github/app

Toutes les issues de github/app

Issues similaires

Plus d'issues AI Infra & Agents

Recevez les nouvelles issues par e-mail

Un résumé court des issues GitHub adaptées aux débutants.