Claude Code 2.1.123 fixes OAuth loop

This reads like a maintenance release on paper, but the auth fix and prompt edits both matter for real-world usage. The CLI is becoming more reliable for locked-down environments, while the prompt changes hint at a broader tradeoff between helpfulness and guardrails.

• –Fixing the OAuth 401 retry loop removes a painful failure mode for teams running with experimental betas disabled.
• –The system prompt change that drops the explicit upfront security-use restriction is notable; it suggests Anthropic is rebalancing how Claude Code handles safety-sensitive requests.
• –Allowing generated or guessed URLs is a quality-of-life improvement, but it also raises the odds of stale or fabricated links slipping into responses.
• –The added and removed `claude-empty-*` model surface entries look like internal plumbing cleanup rather than a user-facing feature.
• –For developers, this is less about new capability and more about trustworthiness, especially if Claude Code is wired into authenticated or policy-constrained workflows.