Custom agent reverts to default; cannot set a custom agent as persistent default
- Dominant language
- No language data
- Stars
- 2.1k
- Forks
- 153
- PR merge metrics
- No merged PRs in 30d
Description
### Summary
Two related issues with **custom agents** in the GitHub Copilot app, surfaced during a customer evaluation (designer-focused UI prototyping workflow, macOS):
1. **Custom agent reverts to "default agent."** After selecting a custom agent for a session, the app reverts the selection back to the default agent (observed across sessions / on relaunch).
2. **No way to set a custom agent as the persistent default.** There is no supported setting to pin a custom agent so that new sessions start with it instead of the default agent. Users must re-select the custom agent every time.
### Impact
For non-technical users (e.g. designers) this "re-select the agent every session" toil is a real adoption blocker — one of the customer's explicit requirements is *reduced manual per-session option-selecting*. Today the Sessions settings expose a default **model**, custom instructions, and auto-approve, but not a default **agent**.
### Environment
- GitHub Copilot app (agent-native desktop), macOS (Apple Silicon)
- Custom agents configured per the custom-agents configuration docs
### Expected
- A selected custom agent persists across sessions/relaunch.
- A supported way to set a custom agent as the persistent default for new sessions (parallel to the existing default-model setting).
### Notes
Filed on behalf of an account team evaluating the app against a customer's requirements. Happy to connect the reporting customer for repro details if useful.
Contributor guide
Research direction
Start with the Sessions settings and custom-agents configuration described in the issue, then trace how the selected agent is stored across sessions and app relaunches. Verify both persistence of a selected custom agent and a setting for choosing the default agent by creating a session, restarting the app, and checking the resulting selection.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- github
- Domain
- ai, desktop
- Issue type
- Feature
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Quiet
- Clarity
- Mostly clear
- Newbie friendliness
- 45/100