gemini-cli: support Login with Google (oauth-personal), mirroring codex auth.json The…
- User
gemini-cli: support Login with Google (oauth-personal), mirroring codex auth.json The
codexagent injects a cached OAuth credential (~/.codex/auth.json) so runs can use a login instead of an API key (CODEX_AUTH_JSON_PATH/CODEX_FORCE_AUTH_JSON), andclaude-codeacceptsCLAUDE_CODE_OAUTH_TOKEN(#173).gemini-clihas no equivalent — it only wiresGEMINI_API_KEY/ Vertex env, so users with a "Login with Google" session (~/.gemini/oauth_creds.json) must keep a separate API key for runs.Proposal: add the same opt-in OAuth-credential injection to the
gemini-cliagent, mirroring the codex pattern —GEMINI_OAUTH_CREDS_PATH/GEMINI_FORCE_OAUTH, default behavior unchanged (fully backward-compatible).I have a working implementation that mirrors the codex auth flow (
_resolve_oauth_creds_path→ upload → chown →~/.gemini) with tests ported 1:1 fromtest_codex_auth.py, verified end-to-end (gemini-cli --ae GEMINI_FORCE_OAUTH=truecompleted hello-world, reward 1.0). Happy to open the PR or hand it off — whichever you prefer.Interface
No new interfaces are introduced.
- Assistant
I’ll trace the existing Codex credential flow and Gemini agent/tests, then implement the matching opt-in behavior and run the repository-required checks.
- Assistant
The existing Codex path uses an explicit opt-in resolver, stages the host credential under
/tmp, fixes ownership, and symlinks it into the CLI home. I’m applying that same lifecycle to Gemini, including best-effort secret cleanup and preserving the default API-key/Vertex behavior. - Assistant
The resolver and staging tests pass. I’m now checking the surrounding Gemini configuration behavior to ensure the injected token also selects the non-interactive
oauth-personalauth mode, rather than merely placing the file where the CLI can find it. - Assistant
The OAuth branch now explicitly selects Gemini CLI’s
oauth-personalmode, so it remains non-interactive even when no API key is present. Formatting, linting, and type checking are clean; I’m running the full unit suite now.
Other models, same or similar prompt
Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.
Identical first prompt
- composer-2.5Compare side by side →
- composer-2.5Compare side by side →
- composer-2.5Compare side by side →
- composer-2.5Compare side by side →
- composer-2.5Compare side by side →
- gpt-5.6-solCompare side by side →
- gpt-5.6-solCompare side by side →
- gpt-5.6-solCompare side by side →
- gpt-5.6-solCompare side by side →
- fable-5Compare side by side →
- fable-5Compare side by side →