Skip to content
cccskills
FreePublished on GitHub

chemistry-research-router

理解化学科研自然语言需求,生成带来源绑定的 ResearchIntent,并通过本地确定性校验路由到七个化学 Skill、受控 Skill 链或 Workflow A/B。用于复杂、多步、模糊、需要自动编排,或可能联网、产生费用和发送数据的化学身份、结构、特征、分子库、反应和已有路线任务;毒性预测、路线生成、实验安全和放大审批不支持。

The English summary is being prepared. Showing the original description.

See how to install

Files included (52)

  • SKILL.md5.4 KB
  • agents/openai.yaml385 B
  • assets/clarification-templates-v1.json1.1 KB
  • references/attachment-manifest-v1.schema.json1.6 KB
  • references/certification-record-v1.schema.json1.4 KB
  • references/clarification-request-v1.schema.json2.3 KB
  • references/research-intent-v1.schema.json10.6 KB
  • references/route-catalog-v1.json13.1 KB
  • references/route-confirmation-v1.schema.json1.5 KB
  • references/route-decision-v1.schema.json5.9 KB
  • references/router-execution-request-v1.schema.json9.2 KB
  • references/routing-boundaries.md1.9 KB
  • references/routing-examples.md1.5 KB
  • scripts/build_intent.py6.4 KB
  • scripts/bundle_install_cli.py1.4 KB
  • scripts/bundle_manifest.py12.9 KB
  • scripts/bundle_spec.py1.4 KB
  • scripts/certification_contract.py2.1 KB
  • scripts/chain_definitions.py7.9 KB
  • scripts/chain_handoffs.py8.9 KB
  • scripts/chain_lock.py1.4 KB
  • scripts/chain_nodes.py11.5 KB
  • scripts/chain_runner.py12.7 KB
  • scripts/chain_validation.py8.1 KB
  • scripts/confirmation_contract.py2.8 KB
  • scripts/decision_contracts.py5.8 KB
  • scripts/direct_preparation.py12.1 KB
  • scripts/direct_runner.py7.9 KB
  • scripts/execution_authorization.py7.4 KB
  • scripts/install_bundle.py13.0 KB
  • scripts/installation_smoke_cases.py5.7 KB
  • scripts/installation_smoke.py7.5 KB
  • scripts/intent_builder.py11.2 KB
  • scripts/policy_guard.py7.2 KB
  • scripts/request_builders.py7.4 KB
  • scripts/request_contracts.py9.8 KB
  • scripts/request_library_builder.py6.2 KB
  • scripts/request_target_builders.py13.1 KB
  • scripts/requirements.txt19 B
  • scripts/route_catalog_spec.py4.9 KB
  • scripts/route_catalog.py8.9 KB
  • scripts/route_engine.py10.3 KB
  • scripts/route_intent.py3.5 KB
  • scripts/router_contracts.py2.5 KB
  • scripts/run_router.py9.5 KB
  • scripts/runtime_layout.py4.7 KB
  • scripts/schema_validation.py3.0 KB
  • scripts/source_binding.py5.3 KB
  • scripts/target_runner.py6.7 KB
  • scripts/target_staging.py2.2 KB
  • scripts/validate_installation.py10.8 KB
  • scripts/validate_intent.py4.1 KB

SKILL.md (original)

Review the instructions the agent will receive before you install.

化学科研确定性路由

将用户的化学科研自然语言目标转换为带来源绑定的 ResearchIntent V1,再交给本地 Validator、Policy Guard、Catalog 和 Runtime。Agent 只负责语义识别,不决定执行命令、科学默认值或自由工作流。

何时使用

  • 任务包含多个化学步骤,或用户要求完整证据链;
  • 目标可能涉及联网、费用、附件外发或特殊科学参数;
  • 用户目标、化学对象、输入 Artifact 或执行范围存在真实歧义;
  • 需要在七个原子 Skill、四条固定 chain 和 Workflow A/B 之间确定唯一入口。

明确、离线、无风险的单步任务可以直接进入对应原子 Skill。边界见 routing-boundaries.md。

Semantic draft

Draft 只包含 Agent 的语义判断和原文证据,固定形状如下:

{
  "schema_version": "1.0.0",
  "language": "zh-CN",
  "goal": {
    "goal_type": "compute_molecular_features",
    "chain_requirement": "explicit_bounded_chain",
    "evidence_text": "用户原文中的完整目标"
  },
  "research_objects": [
    {
      "object_type": "compound_collection",
      "evidence": {
        "source_kind": "attachment",
        "attachment_id": "structures-csv"
      }
    }
  ],
  "requested_operations": [
    {
      "operation_type": "standardize_structure",
      "negated": false,
      "evidence_text": "用户原文中的唯一片段"
    },
    {
      "operation_type": "compute_fingerprint",
      "negated": false,
      "evidence_text": "用户原文中的指纹计算片段"
    }
  ],
  "input_artifacts": [
    {"attachment_id": "structures-csv", "role": "structure_input"}
  ],
  "user_parameters": [],
  "candidate_targets": ["structure-features-v1"],
  "ambiguities": [],
  "unsupported_goals": []
}

消息对象的 evidence 固定为 {"source_kind":"message_span","text":"原文唯一片段"}。操作顺序由数组顺序确定。 参数项只允许 field_id、value、evidence_text。

执行协议

  1. 读取用户原始消息和附件 manifest,保留原文,不 trim、不改换行、不做 Unicode normalization。
  2. 对完整语义做意图识别,只生成紧凑 semantic draft。Agent 负责选择受控语义枚举,并为 goal、操作和显式参数提供原文中唯一出现的 evidence_text;附件对象只引用 attachment ID。
  3. 只记录用户明确给出的科学参数。没有原文证据的参数不得写入 draft;真实缺口进入 ambiguities。
  4. 调用 build_intent.py。构建器生成稳定 ID、精确 span、附件绑定、SHA-256、user_explicit provenance 和 intent_fingerprint,并将 --attachment-root 中通过 hash/size 校验的附件复制到 Intent 同目录供 Request Builder 使用。禁止 Agent 手工生成机械字段或自行 staging。
  5. 调用 run_router.py route 生成 RouteDecision,并在可执行时生成 RouterExecutionRequest。
  6. 按 Decision 状态处理:
    • auto_execute:调用 run_router.py execute;
    • confirmation_required:展示受控原因,得到用户明确确认并生成绑定的 confirmation 后才执行;
    • manual_target_required:仅展示目标,不自动执行;
    • clarification_required:只提出模板指定的问题;
    • unsupported:明确当前能力不支持并停止。
  7. 对 awaiting_human run 使用 run_router.py resume 和绑定当前 gate 的 HumanDecision。
  8. 只报告 Validator、run 状态、Artifact 和 evidence package 中存在的事实。

禁止

  • 禁止用关键词匹配作为主路由;
  • 禁止 Agent 补充科学参数;
  • 禁止绕过 Validator;
  • 禁止自由拼接 Skill;
  • 禁止用 semantic draft 构建器做关键词或正则语义识别;
  • 禁止在 draft 中提供 ID、span、hash、fingerprint、provenance 或命令;
  • 禁止从 Intent 接受 command、entrypoint、Validator path、URL 或凭据;
  • 禁止把 Agent 解释、程序成功或文件生成当作科学结论。

文件接口

python scripts/build_intent.py \
  --draft semantic-draft.json \
  --source source.txt \
  --attachments attachments.json \
  --attachment-root inputs \
  --certificate certificate.json \
  --intent intent.json

python scripts/run_router.py route \
  --intent intent.json \
  --source source.txt \
  --attachments attachments.json \
  --certificate certificate.json \
  --decision decision.json \
  --request execution-request.json

python scripts/run_router.py execute \
  --request execution-request.json \
  --decision decision.json \
  --run-dir router-run \
  --installation-receipt .chemistry-agent-bundle/installation-receipt.json

python scripts/run_router.py resume \
  --run-dir router-run \
  --decision human-decision.json \
  --installation-receipt .chemistry-agent-bundle/installation-receipt.json

教学用语义例子见 routing-examples.md。不得将例子或历史测试标签写入用户请求。

Reviews

No reviews yet. Tell others what you thought.

Comments

No login needed. Share how it worked for you, ask a question, or note where it did not run.

No comments yet. Be the first to write one.

Comments are published right away. Do not include personal information such as email addresses or phone numbers. Up to 2 URLs.

0 / 1,000 characters

This form is protected by reCAPTCHA, and the Google Privacy Policy and Terms of Service apply.

More skills from this repository

中文购车决策助手初始化与验收 Skill。用于安装、配置、启动、恢复或诊断仓库内完整前后端,引导用户选择自己的 AI Native 应用开发底座 Workspace,私密配置 Agent Plan Key,真实调用专业数据集核验 1–3 款候选车型、配置与城市车系月度数据,并完成应用开发底座和浏览器验收。用户提出“初始化购车决策助手”“安装选车网站”“配置汽车专业数据”“启动购车助手”或排查该应用时使用。

The English summary is being prepared. Showing the original description.

volcengine/ai-app-lab2.5KUpdated Sep 7, 2026

对已标准化化合物确定性计算受控二维描述符、Morgan/RDKit/MACCS 指纹和数据集质量画像。用于准备分子特征、生成指纹,或检查特征缺失、常数、异常和分布。

The English summary is being prepared. Showing the original description.

volcengine/ai-app-lab2.5KUpdated Sep 7, 2026

整理和审查结构化化学反应,保留原始记录并检查参与物、角色、产率、重复、守恒和 ORD 合同。用于清洗 reaction SMILES、ORD 或反应表格。

The English summary is being prepared. Showing the original description.

volcengine/ai-app-lab2.5KUpdated Sep 7, 2026

中文个人投资助手初始化 Skill。引导用户确认真实股票、市场、关注偏好和盘后监控安排,检查 Agent Plan Harness,安装并启动仓库内完整前后端,导入个性化配置,真实调用 DataPro、豆包搜索和 Agent Plan 模型生成首批个股简评与盘后风险摘要,并完成可用性验收。用户提出“帮我初始化个人投资助手”“安装投资助手”“配置我的关注股票”“让我打开网站就有内容”或需要恢复、诊断该应用时使用。

The English summary is being prepared. Showing the original description.

volcengine/ai-app-lab2.5KUpdated Sep 7, 2026

保守解析化学名称、SMILES、InChI、InChIKey、PubChem CID、ChEMBL ID 或 CAS RN,保留多源证据、相关形式、歧义和冲突。用于确认化学记录是谁、对齐开放数据库,或在结构标准化前选择候选。

The English summary is being prepared. Showing the original description.

volcengine/ai-app-lab2.5KUpdated Sep 7, 2026

评审已有合成路线的拓扑、逐步反应先例、库存声明和证据缺口。用于比较 AiZynthFinder、PaRoutes 或结构化多步路线并生成专家复核队列。

The English summary is being prepared. Showing the original description.

volcengine/ai-app-lab2.5KUpdated Sep 7, 2026

See all skills by volcengine

Report a problem with this skill