Github Copilot consuming more than one premium requests in a single session

[his-majest]

Description

I am using oh-my-opencode. In a session, when main agent (premium request consumer) comes back after background task completion or after compaction, it is a consuming a premium request. Subagents are not consuming premium requests so far. Basically, it is consuming requests while switching agents.

Plugins

oh-my-opencode

OpenCode version

1.1.48

Steps to reproduce

No response

Screenshot and/or share link

No response

Operating System

Linux Mint 22.2

Terminal

Xfce Terminal

[github-actions]

This issue might be a duplicate of existing issues. Please check:

• Copilot auth now sets far too many requests as "user" consuming premium requests rapidly #8030: Copilot auth now sets far too many requests as "user" consuming premium requests rapidly - describes synthetic messages being marked as user-initiated instead of agent-initiated
• GH Copilot | synthetic user messages burn premium requests #8700: GH Copilot | synthetic user messages burn premium requests - specifically about subagent messages consuming premium requests
• Github Copilot multiple premium request charges without subagents #8067: Github Copilot multiple premium request charges without subagents - reports multiple premium charges in same session
• more than one copilot credit is used for tasks #10572: more than one copilot credit is used for tasks - describes tasks consuming multiple credits

Feel free to ignore if your case differs from these.

[mykola-dev]

Yeah, I noticed that too. OpenCode burns premium requests like crazy, but with Copilot CLI I can vibecode for hours and barely touch my limit.

[ayaka209]

same here

[his-majest]

Only thing is oh-my-opencode is only available for opencode, otherwise I was using GitHub copilot.
oh-my-opencode agents are crazy, if someone could build similar thing for copilot, it would be great

[paolomainardi]

I haven't investigated yet, but can the reasons be found regarding the models used to run tools, such as listing files and reading? As far as I know, Claude Code (and possibly Copilot) uses smaller models for those tasks.

[utapyngo]

Each compaction uses 2 premium requests in opencode. In Copilot CLI and Copilot Chat in vscode, compaction does not use premium requests since x-initiator is set to agent for auto-compaction.

[georgeharker]

I'd love to get some feedback on

• whether copilot agreed that post-compaction responses could be agent rather than user
• is there a bug in compaction because it appears it kicks in super aggressively compared to the point at which i would run out of context for a comparable compaction-disabled session.

The combination drastically increases the request burn as each compaction fires 2x premium requests (minimum one if you use a small free model for compaction). And can do so many many times during a session without user intervention.