My Claude Code Sign-in Fix Needed Two Fixes of Its Own

Description and use

Refmade’s sign‑in logic mistakenly allowed API‑key sign‑ins in readiness checks but not in actual runs, causing failures. After correcting the sign‑in handling and addressing regressions, all tests passed and real team runs succeeded.

Problem

The Refmade desktop app refused sign-ins that the Claude Code CLI accepts, causing readiness checks to report success while actual runs failed or silently switched accounts.

Outcome

After the fixes, all automated tests passed and real team runs completed successfully, demonstrating that the app’s readiness checks and actual runs now use consistent configuration.

Workflow

• 1. Removed the refusal of API‑key sign‑ins on September 29.
• 2. Ran an adversarial review on September 30, which discovered the mismatch between readiness check and run.
• 3. Reproduced the issue with a fake API key and local test server.
• 4. Updated the sign‑in logic to pass the settings file only when the sign‑in comes from it, using environment checks to distinguish keys.
• 5. Fixed additional regressions: removed forced provider for Codex, limited environment variables per tool, corrected login labeling.
• 6. Ran comprehensive tests (405 app, 323 CLI, 401 engine) and mutation checks.
• 7. Verified real team runs completed in 6.1 s (Claude) and 14.3 s (Codex).

Evidence

The app has a "team" mode, where several agents plan, write and check a task. Until September 29 it only worked when Claude Code was signed in with a Claude account (the plan login). If you had an API key instead, the app answered with a reason code, AUTH_MODE_UNSUPPORTED, and stopped.

Source

Team mode starts Claude Code with --restricted, which keeps settings files out of the run. That includes your own user settings file. But before a run, the app asks claude auth status --json whether you are ready, and that command does read the user settings file.

Source

After the fixes: app tests 405, CLI tests 323, engine tests 401 plus one. I also ran mutation checks, which deliberately break the code and confirm a test fails. Real team runs through the actual CLIs finished in 6.1 seconds for Claude and 14.3 seconds for Codex.

Source