lsm / neokai / 27735749761

82%

Build Type

push

github

Committed by web-flow

Commit Message

refactor: remove Codex SDK aliasing, use real model IDs (#2130)

* refactor: remove Codex SDK aliasing, use real model IDs

Following the GLM/Kimi pattern from task #606, Codex models now use real
model IDs directly instead of aliasing to Anthropic SDK IDs.

Changes:
- Removed CODEX_TO_SDK_ANTHROPIC_MODEL mapping (replaced with CODEX_TO_SDK_MODEL identity mapping)
- Removed SDK_ANTHROPIC_TO_CODEX_MODEL reverse mapping (no longer needed)
- translateModelIdForSdk() now returns 'default' (matches GLM/Kimi pattern)
- ANTHROPIC_DEFAULT_*_MODEL env vars now use real Codex IDs (e.g., 'gpt-5.5' instead of 'claude-opus-4-7')
- Removed Anthropic alias models from bridge /v1/models response
- Updated all tests to expect real Codex IDs

Benefits:
- Removes alias indirection complexity
- Ensures SDK uses correct 272k context window from /v1/models metadata
- Avoids token counting mismatch from wrong model ID
- Matches proven GLM/Kimi pattern

Context window is read from /v1/models metadata (preferContextWindowMetadata: true)
instead of SDK's hardcoded database, eliminating the need for Anthropic aliases.

* docs: clean up orphaned comments referencing removed Codex aliasing

Remove stale documentation and test fixtures that still reference the removed
`claude-opus-4-7` / `claude-sonnet-4-20250514` aliasing pattern, as requested
in PR review.

Changes:
- openai-responses-bridge/server.ts: Updated interface docs for setSessionModelConfig
  to reflect that Codex now uses identity mapping
- model-service.ts: Updated step-5 comment about sdkModelIds
- query-options-builder.ts: Updated NATIVE_CONTEXT_WINDOW_PROVIDER_IDS doc for Codex
- context-fetcher.ts: Updated capacity-mismatch comment (no longer relevant)
- Test fixtures: Updated 5 test files to use real Codex IDs instead of Anthropic aliases

All tests pass (61 openai-responses-bridge + 75 context-fetcher/model-service).

Coverage Stats

10722 of 15627 branches covered (68.61%)

Branch coverage included in aggregate %.

19 of 19 new or added lines in 2 files covered. (100.0%)

1 existing line in 1 file now uncovered.

91394 of 109312 relevant lines covered (83.61%)

297.12 hits per line