Deep research — tumor immunotherapy survey driven by a Chinese prompt

PASS · rubric SATISFIED649sprompt: zhartifact: zh

Prompt (verbatim)

no-interview light-research: 使用 bioresearcher-deep-research 技能，帮我做一个关于肿瘤免疫治疗（tumor immunotherapy）的文献综述并附引用。最多 2 个研究方面，每个方面的工作节点最多调用 10 次 biomcp 工具。请用中文撰写报告，引用使用 PMID。

What happened

Chinese prompt (the WorkBuddy connector's canonical zh example, extended with an explicit call budget): queried English sources, reported in Chinese.
Two Chinese-titled parallel workers (ICB literature worker + CAR-T literature worker) built per-aspect evidence ledgers; render numbered 52 citations from the merged ledger and vet-references passed its structural audit.
Objective check asserted Chinese text in the final answer (regex [\u4e00-\u9fff]) — bilingual behavior is verified, not claimed.

Evidence

The evidence below is in Chinese — a true-run output, untranslated.

Embedded report artifactOpen full report ↗

final_report.md · screenshot (chromium capture)

Objective checks

checks passed: 8/8 · one rubric flag adjudicated SATISFIED post-run (recorded in result.json)

Fan-out mode	Parallel fan-out via the harness Task tool (2 Chinese-titled subagent workers captured live)
Provenance	`d818c85` · opencode 1.18.30 · biomcp@1.4.0

Replay

node demos/run-demo.mjs --only agent-deep-research-zh --publish

Manual-run only — this replay spends LLM tokens (real opencode CLI + network).

In the repository: README · transcript · result.json · artifact directory (permalink)