Summary

• Add Claude-facing aliases for Opus 4.7 and 1M Sonnet/Opus variants, with local token-limit entries for the 1M aliases.
• Add Codex-facing gpt-5.5, gpt-5.5-*, gpt-5.4, and explicit Codex model mappings.
• Accept xhigh/max reasoning efforts in OpenAI-compatible request models, map Responses reasoning effort into Anthropic thinking budgets, and apply Codex model-alias effort defaults.
• Normalize Codex-compatible string input payloads, clamp max to xhigh before Codex backend dispatch, preserve detected Codex CLI version in fallback headers, and allow x-api-key as a shared-secret header for compatible clients.

This is intentionally parallel to #64 and #65. It does not depend on Anthropic tool-history compaction or the Responses fc_* item-id fix.

For 1M Claude aliases, this PR exposes aliases and token limits while relying on the existing Anthropic beta-header merge behavior to preserve client- or CLI-provided long-context beta tags. It does not force long-context beta headers for every account/model combination.

Validation

• ./Taskfile check
• uv run pytest tests/unit/auth/test_auth.py tests/unit/llms/formatters/test_openai_to_anthropic_helpers.py tests/unit/llms/formatters/test_openai_to_openai_reasoning.py tests/plugins/codex/unit/test_adapter.py tests/unit/plugins/test_codex_detection.py tests/unit/utils/test_model_mapper.py tests/plugins/max_tokens/test_service.py tests/plugins/claude_api/integration/test_claude_api_basic.py -q - 115 passed
• uv run pytest tests/unit tests/plugins/codex/unit/test_adapter.py tests/plugins/max_tokens/test_service.py tests/plugins/claude_api/integration/test_claude_api_basic.py -q - 671 passed
• git diff --check