-
Notifications
You must be signed in to change notification settings - Fork 295
Description
Problem
When using the assign-to-agent safe output to programmatically assign the Copilot coding agent to an issue, there is no way to specify which model the coding agent should use. The assignment always defaults to "Auto" model selection, which currently only includes Claude Sonnet 4.5.
This makes automated workflows unable to control the model used by the coding agent, even though the GitHub UI allows manual model selection when assigning Copilot to an issue (Auto, Claude Sonnet 4.5, Claude Opus 4.5, Claude Opus 4.6, GPT-5.1-Codex-Max, GPT-5.2-Codex).
Current behavior
The assign-to-agent safe output only accepts issue_number (or pull_number) and agent parameters:
assign-to-agent:
agent: copilot
issue_number: 107The coding agent always launches with "Auto" (= Claude Sonnet 4.5).
Desired behavior
Add an optional model parameter to assign-to-agent:
assign-to-agent:
agent: copilot
issue_number: 107
model: claude-opus-4.6 # optional, defaults to "auto" if omittedAlternatively, support a repository-level or workflow-level default model setting (e.g., a repo variable like GH_AW_COPILOT_AGENT_MODEL) that the assign-to-agent output would use.
Use case
I have an orchestrator workflow that scans issues, determines what needs work, and uses assign-to-agent to dispatch the Copilot coding agent. However, some tasks require a more capable model like Claude Opus 4.6, and every programmatic assignment launches with Sonnet 4.5 instead. This forces me to manually cancel and re-assign each task through the GitHub UI to select the right model, which defeats the purpose of automation.
Workaround
Currently the only workaround is to not use assign-to-agent and instead have the workflow post a comment notifying a human to manually assign the coding agent with the desired model through the GitHub UI.