fix(core): inject thinking blocks for DeepSeek anthropic-compatible provider#3788

Merged

wenshao merged 16 commits into

mainfrom

fix/deepseek-anthropic-thinking

May 2, 2026

Commits on May 2, 2026

fix(core): inject thinking blocks for DeepSeek anthropic-compatible provider
wenshao
committed
fix(core): gate DeepSeek thinking-block injection on thinking mode
wenshao
committed
fix(core): narrow DeepSeek thinking injection to tool_use turns + subdomain test
wenshao
committed
docs(core): align DeepSeek thinking-injection comments with narrowed scope
wenshao
committed
fix(core): per-request DeepSeek detection + strip thinking when off
wenshao
committed
fix(core): preserve thinking-only assistant turns instead of emitting empty content
wenshao
committed
fix(core): validate thinking-block signature, rename option, gate output_config
wenshao
committed
fix(core): clean up non-compliant thinking blocks on plain-text turns + assert output_config gating
wenshao
committed
test(core): add explicit redacted_thinking injection-path coverage
wenshao
committed
fix(core): per-request anthropic-beta + normalize thinking-only turns
wenshao
committed
fix(core): preserve thinking text by normalizing in place + merge user beta flags
wenshao
committed
docs(core): align thinking-injection comments with normalize semantics + add stream test
wenshao
committed
refactor(core): split DeepSeek thinking option in two + add header coexistence test
wenshao
committed
fix(core): case-insensitive customHeaders[anthropic-beta] merge
wenshao
committed
docs(core): align thinking-injection docs with normalize-in-place semantics + redacted_thinking strip test
wenshao
committed
fix(core): exclude anthropic-beta variants from defaultHeaders to avoid wire duplication
wenshao
committed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

fix(core): inject thinking blocks for DeepSeek anthropic-compatible provider#3788

fix(core): inject thinking blocks for DeepSeek anthropic-compatible provider#3788
wenshao merged 16 commits into
mainfrom
fix/deepseek-anthropic-thinking

Commits on May 2, 2026

fix(core): inject thinking blocks for DeepSeek anthropic-compatible provider

fix(core): gate DeepSeek thinking-block injection on thinking mode

fix(core): narrow DeepSeek thinking injection to tool_use turns + subdomain test

docs(core): align DeepSeek thinking-injection comments with narrowed scope

fix(core): per-request DeepSeek detection + strip thinking when off

fix(core): preserve thinking-only assistant turns instead of emitting empty content

fix(core): validate thinking-block signature, rename option, gate output_config

fix(core): clean up non-compliant thinking blocks on plain-text turns + assert output_config gating

test(core): add explicit redacted_thinking injection-path coverage

fix(core): per-request anthropic-beta + normalize thinking-only turns

fix(core): preserve thinking text by normalizing in place + merge user beta flags

docs(core): align thinking-injection comments with normalize semantics + add stream test

refactor(core): split DeepSeek thinking option in two + add header coexistence test

fix(core): case-insensitive customHeaders[anthropic-beta] merge

docs(core): align thinking-injection docs with normalize-in-place semantics + redacted_thinking strip test

fix(core): exclude anthropic-beta variants from defaultHeaders to avoid wire duplication