[ranxianglei]

Release v0.1.38

Supersedes 0.1.37 (which had no changelog entry — backfilled here).

0.1.38

• OpenAI reasoning_content round-trip for thinking-mode models (fix(openai): preserve reasoning_content round-trip for thinking-mode models #129): DeepSeek-R1 / GLM-4.6 thinking / Qwen-QwQ emit reasoning_content (CoT) and require echo-back. bili's OpenAI adapter dropped it → HTTP 400 The reasoning_content in the thinking mode must be passed back to the API. on every re-request. Now round-tripped end-to-end (openaiToCore/coreToOpenai + streaming + re-request reconstruction), mirroring the existing Anthropic thinking handling. Also fixes double-forwarding when a chunk carries both reasoning_content + content.

0.1.37 (backfilled)

• MITM domain auto-discovery (feat(mitm): auto-discover model domains from client configs (no hardcoding) #125): bili reads client configs and auto-builds the MITM whitelist from discovered provider hostnames. open.bigmodel.cn / zcode.z.ai / api.z.ai are now discovered, not hardcoded; defaults reduced to the three binary-baked endpoints.
• Anthropic round-2 streaming framing / vertical-text bug (fix(anthropic): forward round-2 stream with proper framing (vertical-text bug) #126): round-2 text after a proxy tool/compress re-request rendered as vertical text. Now carries correct content_block_start/delta/stop framing.
• HTTP-proxy-mode absolute URLs (fix(forward): handle absolute URLs in HTTP-proxy mode #127): bili mangled absolute URLs (https://api.anthropic.comhttp://127.0.0.1:18081/…). Absolute URLs now forwarded to their host.

---

• release v0.1.38 (package.json + lockfile) + docs: changelog
• 343/343 tests, typecheck clean, build OK.
• acp-kernel unchanged at 0.0.19.