fix(agent): add Chinese reasoning tags to think-block filter for MiniMax M3

[liuhao1024]

Problem

MiniMax M3 emits Chinese-language reasoning tags ( 思考, 反思, 推理, 推敲) instead of the English <think> family. The three filter sites that strip reasoning content only handle English tags, causing raw Chinese reasoning to leak into user-visible output on Telegram, Discord, and CLI.

Observed symptom: users see fragments like 思考\n...raw reasoning... in the middle of assistant responses.

Solution

Add the four Chinese reasoning tag pairs to all three filter sites:

#	File	Symbol	Change
1	gateway/stream_consumer.py	_OPEN_THINK_TAGS / _CLOSE_THINK_TAGS	+4 tags each
2	cli.py	_stream_delta local _OPEN_TAGS / _CLOSE_TAGS	+4 tags each
3	agent/agent_runtime_helpers.py	strip_think_blocks()	+4 closed-pair regexes, +1 unterminated-block regex, +1 stray-orphan regex

Testing

• 12/12 new regression tests pass (test_stream_consumer_chinese_think_tags.py)
• 93/93 existing stream_consumer tests pass (no regression)
• Tests cover: closed pairs for all 4 tag types, unterminated blocks, stray orphans, mixed English+Chinese, prose mentions not over-stripped

Related Issues

• Fixes [Bug]: Chinese reasoning tags ( 思考 / 反思 / 推理 / 推敲) from MiniMax-M3 leak to user output in streaming + final response #43827
• Related: [Bug] MiniMax reasoning content leaks to CLI output even with reasoning_effort: none #17924 (English tags on CLI), Streaming reasoning tag filters miss mixed-case tag variants #27288 (mixed-case tags in streaming)

[teknium1]

Thanks for this, and for the detailed analysis in the linked issue. We're going to close it though — it runs against a standing design line: Hermes does not add parsing/scrubbing layers to strip a given model's bespoke raw reasoning delimiters.

The existing English <think> family scrubbers are the boundary we hold; we don't extend that surface to chase each model's custom reasoning markers (思考/反思/推理/推敲, mm:think, and whatever the next model emits). That set grows without end, and it masks the real issue: reasoning that leaks inline is a reasoning-routing / endpoint problem, not a missing-filter problem.

Concretely for MiniMax-M3: our provider plugin already sends reasoning_split: True to the api.minimax.io/v1 OpenAI-compatible route, so the model returns reasoning in a separate reasoning_content field rather than inline. If reasoning is still reaching delta.content, that points at the reasoning-split routing for that deployment, not at a tag the scrubber should learn.

There's also a correctness hazard specific to this fix: 思考/推理/反思 are ordinary Chinese words, not XML-shaped sentinels. Stripping everything from a bare 思考 (or between two occurrences of it) would corrupt legitimate Chinese-language answers — a fix that damages the feature it's meant to protect.

Not a knock on the work — the leak you saw is real to you. But the right fix lives in reasoning routing, not in a new raw-reasoning parser. Closing.