Codex Login Manager v1.0.0-beta.2

Codex Login Manager diagnoses Codex CLI authentication on macOS and starts the official ChatGPT OAuth login flow.

This community build is open source and ad-hoc signed, but it is not Apple-notarized yet. Try to open it once; if macOS blocks it, open System Settings > Privacy & Security, click Open Anyway for Codex Login Manager, and confirm Open. Never disable Gatekeeper globally.

The app reads ~/.codex/auth.json locally, invokes the installed official codex CLI, redacts secrets before showing command output, and does not upload credentials or write to config.toml.

Changes in v1.0.0-beta.2

Security

• Hardened named-secret redaction for quoted and escaped JSON fields, camelCase
  keys, authorization headers, bearer-token fields, OAuth client secrets, and
  opaque values wrapped across structurally bounded, indented lines.
• Hardened API-key and JWT redaction when token fragments are wrapped across
  adjacent indented lines. Ambiguous token-shaped continuation lines are now
  conservatively redacted, while blank lines, prose, structured diagnostics,
  and indentation rollbacks stop the scan.
• Redacted three-segment JWTs, including unsecured JWTs with an empty signature,
  and five-segment compact JWEs with the empty encrypted-key segment used by
  direct encryption.
• Rejected placeholder-like values and trailing secret text instead of trusting
  any value that merely starts with a redaction marker.
• Required a successful CLI exit and an exact positive status line before
  reporting ChatGPT login, so negated or failing output cannot spoof success.
• Stopped reflecting unrecognized auth_mode values from auth.json into the
  interface; unsupported modes now use fixed localized text.
• Preserved surrounding JSON fields and diagnostic structure instead of
  discarding internal indentation and blank lines during sanitization.

Tests

• Added adversarial coverage for quoted and escaped JSON, authorization headers,
  compact JWE, multiline secrets, CRLF output, malformed token shapes,
  identifier-boundary false positives, and idempotence.
• Added service-level tests using temporary Codex CLI executables to verify that
  stdout and stderr are both scrubbed, unfamiliar CLI text remains visible,
  malformed auth.json contents are never echoed, and unrecognized
  auth_mode values cannot reach visible status fields.
• Added status-parsing coverage proving that negated text and nonzero exits do
  not report a successful ChatGPT login.