[diegosouzapw]

Why

Claude Opus 4.7 and later (Opus 4.7/4.8, Fable 5) changed the Messages API request contract — and OmniRoute (where Claude is the most-used provider) still sent the old shapes, so well-formed requests could hard-400:

1. Manual extended thinking is gone. thinking.type:"enabled" or any thinking.budget_tokens now returns 400. Reasoning is adaptive-only, steered by output_config.effort. (Migration guide, extended thinking, oh-my-openagent#4614). The OpenAI→Claude translator mapped reasoning_effort low/medium/high to a manual thinking:{type:"enabled", budget_tokens} → 400 on those models.
2. Sampling params are fixed. Non-default temperature/top_p/top_k return 400. The translator forwarded client-supplied temperature/top_p unconditionally, and the Claude registry models carried no unsupportedParams.

output_config.effort (low/medium/high/xhigh/max) was already correct — no change there.

What

• modelSpecs.ts — new adaptiveThinkingOnly flag + isAdaptiveThinkingOnly() helper, set on claude-opus-4-7/claude-opus-4-8/claude-fable-5 (resolves dated/Bedrock aliases).
• openai-to-claude.ts — for adaptive-only models, reasoning_effort of every level → {type:"adaptive"} + output_config.effort (preserves the requested level), never a manual budget. Pre-4.7 models keep the budget buckets.
• chatCore.ts — normalizeClaudeAdaptiveThinking catch-all at the existing post-translation thinking-normalization chokepoint (next to normalizeThinkingForModel/normalizeMimoThinking): collapses any residual manual thinking that reached dispatch (Claude-native passthrough legacy shape, per-model defaults) to {type:"adaptive"}, keyed on the resolved upstream model so it covers every routing mode. New pure leaf open-sse/services/claudeAdaptiveThinking.ts.
• registry (claude/anthropic) — unsupportedParams: ["temperature","top_p","top_k"] on Opus 4.7+/Fable 5 ids (both dashed claude-opus-4-8 and dotted claude-opus-4.7), stripped at the existing getUnsupportedParams chokepoint.

Scope guard: pre-4.7 models (Opus 4.6/4.5, Sonnet 4.5/4.6, Haiku 4.5) still accept manual budgets and sampling params — regression-tested.

Tests (TDD)

• claude-adaptive-thinking-normalize.test.ts (11) — enabled+budget → adaptive, budget/max_tokens stripped, adaptive untouched, disabled untouched, pre-4.7 budget preserved, alias resolution, no-op identity.
• claude-adaptive-sampling-params.test.ts (5) — sampling strip for Opus 4.7+/Fable 5 (both id shapes) + pre-4.7 regression guard + isAdaptiveThinkingOnly flags.
• translator-openai-to-claude.test.ts — new adaptive-only-by-level test + pre-4.7 budget regression; repointed the Support Claude Effort Level #627 128k-cap regression onto Opus 4.6 (still budget-based).

typecheck:core ✓ · eslint ✓ · check:file-size ✓ (chatCore baseline 5095→5102, justified) · check:provider-consistency ✓ · check:env-doc-sync ✓ · broad Claude/thinking cluster 118/118.

🤖 Generated with Claude Code

[gemini-code-assist]

Warning

You have reached your daily quota limit. Please wait up to 24 hours and I will start processing your requests again!

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.