| name | cross-domain-idea-studio |
| description | Develop evidence-bounded cross-domain research directions by retrieving bridge evidence, extracting transferable mechanisms, auditing migration risk, and submitting candidates to a human selection gate before hypothesis compilation. Use when a researcher wants a defensible cross-domain idea rather than a loose analogy. |
| tools | ["read_file","write_file","write_structured_file","list_files","grep_search","multi_source_search","openalex_search","semantic_scholar_search","crossref_search","fetch_paper_pdf","extract_pdf_text","extract_paper_sections","save_paper_note","compute_idea_novelty_signal","analyze_idea_concentration","ask_human","update_skill_workflow","finish_task"] |
| strict_tools | true |
| model_tier | standard |
| temperature | 0.35 |
| allowed_read_prefixes | ["project.yaml","user_inputs/cross-domain-idea-studio/","user_seeds/","literature/","ideation/","drafts/survey/","_runtime/skill_sessions/"] |
| allowed_write_prefixes | ["user_inputs/cross-domain-idea-studio/","literature/bridge_notes/","literature/cross_domain_catalogs/","ideation/"] |
| outputs_expected | {"bridge_plan":"ideation/cross_domain_bridge_plan.json","evidence_matrix":"ideation/cross_domain_evidence_matrix.md","transfer_cards":"ideation/cross_domain_transfer_cards.md","candidate_pool":"ideation/cross_domain_candidate_pool.json","risk_register":"ideation/cross_domain_risk_register.md","selection":"ideation/cross_domain_selection.json","workflow_manifest":"ideation/cross_domain_manifest.json"} |
| interaction | {"mode":"guided","language":"zh-CN","summary":"从目标问题出发,检索/读取桥接领域证据,提炼可迁移机制、检查失配风险,生成跨域候选并在人工选择后交给 hypothesis compiler。","request_required":true,"request_prompt":"请说明目标问题、已知目标域证据、拟探索或避免的桥接领域、约束和希望的创新类型。","example_request":"用 organizational learning 的机制启发 LLM agent memory 研究;先核验迁移条件,再生成可证伪的跨域候选。","required_inputs":[{"id":"target_problem","label":"目标问题与约束","description":"说明目标研究对象、痛点、现有证据、不可违反的边界和期望的贡献;不能只给一个领域名。","paths":"[Truncated]","extensions":"[Truncated]","min_bytes":80,"example":"user_inputs/cross-domain-idea-studio/target_problem.md"}],"optional_inputs":[{"id":"target_evidence","label":"目标域综合或文献卡","description":"可选;提供后所有候选需优先链接其具体机制、张力和边界。缺失时会被标为 preliminary 并请求补检/补料。","paths":"[Truncated]","extensions":"[Truncated]","min_bytes":40,"example":"literature/synthesis.md"}],"outputs":[{"id":"bridge_plan","label":"跨域桥接检索计划","path":"ideation/cross_domain_bridge_plan.json","description":"源域、目标域、检索问题、迁移条件和阅读优先级。"},{"id":"evidence_matrix","label":"跨域证据矩阵","path":"ideation/cross_domain_evidence_matrix.md","description":"源域机制、论文/section 锚点、目标映射、证据等级和不匹配。"},{"id":"transfer_cards","label":"可迁移机制卡","path":"ideation/cross_domain_transfer_cards.md","description":"每张卡说明源机制、目标映射、迁移风险、替代解释和需要的区分实验。"},{"id":"candidate_pool","label":"跨域候选池","path":"ideation/cross_domain_candidate_pool.json","description":"候选方向、来源、证据、由模型写出的 2-3 条可证伪候选假设、评分边界和选择状态。"},{"id":"risk_register","label":"迁移风险登记册","path":"ideation/cross_domain_risk_register.md","description":"语义失配、单位不一致、因果过度迁移、资源缺口和不能主张的内容。"},{"id":"selection","label":"人工选择记录","path":"ideation/cross_domain_selection.json","description":"用户选择、合并、重构或要求继续补证据的结果;未选择不得冒充 hypothesis handoff。"},{"id":"workflow_manifest","label":"跨域 Idea 工作流清单","path":"ideation/cross_domain_manifest.json","description":"检索、笔记、证据状态、候选 lineage、Gate 和可恢复点。"}]} |
| workflow | {"kind":"integrated","summary":"先证明桥接机制值得讨论,再生成受证据约束的跨域候选;跨域不是唯一 Idea 来源,也不等于可迁移结论。","phases":[{"id":"target_contract","label":"目标问题与目标域证据","objective":"固定目标问题、目标域证据、约束和需要解决的机制张力。","operations":"[Truncated]"},{"id":"bridge_retrieval","label":"Bridge 计划与定向检索","objective":"与用户确认桥接方向后,建立小而可解释的 bridge query portfolio 和来源记录。","operations":"[Truncated]","human_gate":true},{"id":"transfer_audit","label":"迁移机制与失配审计","objective":"从桥接来源提取机制、条件、反例和目标映射,不把类比写成证据。","operations":"[Truncated]"},{"id":"candidate_jury","label":"候选治理与人工选择","objective":"生成可证伪候选、评分范围和 unsupported 项,并让用户选择/合并/重构。","operations":"[Truncated]","human_gate":true}]} |
Cross-domain Idea Studio
Treat the target domain and bridge domain symmetrically as evidence problems. A bridge paper may inspire a candidate but does not prove target-domain validity. For each transfer card, state the source mechanism, source anchor, target mapping, mismatch, evidence level, alternative explanation, and minimum discriminating observation.
Use only the canonical Paper Note roots when existing ResearchOS notes are relevant:
literature/deep_read_notes/ for mainline full/partial reading, literature/bridge_notes/ for bridge full/partial reading, and literature/shallow_read_notes/ for abstract-level recall. Do not read or create retired paper_notes* directories. Shallow notes may surface a Bridge lead or required reading upgrade, but cannot establish a transfer mechanism or target-domain validity.
Use literature/cross_domain_catalogs/ separately for the user-confirmed B1–B# tracks.
Its bridge_context.json and paper_catalog.json preserve retrieved metadata and abstract-level transfer fuel even when no Bridge paper has been deeply read. They can generate a leap, counterexample, mechanism challenge, or validation question, but must remain explicitly conjectural until a canonical note or other verifiable source supports the concrete claim.
Generate candidates only after the transfer audit. When target or bridge evidence is missing, output preliminary, needs_retrieval, or unsupported rather than a scored novelty claim. Never fabricate a current-project dataset, baseline, metric, command, budget, performance estimate, AUUC/Qini value, or experimental result. Concrete details are allowed only when allowed artifacts explicitly source them.
The final selection file must preserve the human decision. Do not claim that a selected candidate has entered hypothesis compilation unless the decision says so; recommend the separate hypothesis-compiler Skill as the next explicit action.
Native T4 handoff boundary
The files produced here are Bridge evidence and optional Candidate Seeds. They are not native T4 Population artifacts: do not write ideation/evidence/, ideation/populations/,
ideation/evolution/, ideation/portfolio.json, or ideation/final_cards/. When the researcher wants these Seeds considered with other routes, hand them to the T4 Evidence Routing flow. It preserves the Bridge escape hatch, decides whether the evidence permits a Candidate, independently scores any resulting Candidate, and keeps unsupported transfers visible rather than forcing a merge. A selected native Candidate produces a Pre-Novelty brief first; formal hypotheses and an experiment plan remain a T4.5-only outcome.