This optional Linux feature patches Codex webview bundles so Copilot-auth sessions can select the reasoning efforts supported by the chosen model.
By default, upstream collapses Copilot model reasoning effort choices to one
medium entry and disables the composer effort controls for Copilot auth. This
feature keeps the UI changes local and opt-in instead of shipping them as a
core Linux compatibility patch. The current upstream app owns persistence via
its model settings service, so the feature does not patch a private storage
key.
Enable it by copying linux-features/features.example.json to
linux-features/features.json and adding the feature id:
{
"enabled": [
"copilot-reasoning-effort"
]
}Then rerun the install or package build so the ASAR patch step can apply the feature to the generated app.
- The semantic model-list owner keeps the model's full
supportedReasoningEffortslist for Copilot auth. - The current dropdown and slash-command owners keep the reasoning effort
controls and
/reasoningcommand enabled when the normal model and effort prerequisites are present. - Every copy of the upstream model settings hook (rspack copies it into each composer surface chunk) is checked, without edits, to still persist the selection through the upstream model settings service.
The rspack webview has no stable chunk names, so each owner is found by content (message ids, settings keys and model-list strings), not by file name.
Run the feature tests with:
node --test linux-features/copilot-reasoning-effort/test.jsOr run all feature tests with:
node --test linux-features/*/test.jsThe patch is fail-soft. If the upstream minified bundle shape changes, the build logs a warning and leaves the affected bundle unchanged.