Catalyst Design

When to use

Users asking how to design a catalyst, which composition to pick, how to optimize performance, synthesis-route advice, catalyst formulation, electrocatalyst design, or how to choose synthesis conditions.
Typical input: design goal (reaction type + material system / performance need). Optional: catalyst-search literature results.
Covers HER, OER, ORR, PEMWE, overall water splitting, photocatalysis, and plastic upcycling, among other systems.

When NOT to use

Pure literature search — use catalyst-search; this skill does not re-search.
Non-catalysis material design (e.g., battery cathodes, semiconductor devices) — not applicable.
Looking up a specific DOI or full text — use WebSearch / WebFetch directly.

Prerequisites

Ships with its own knowledge base (references/) and templates (templates/); runs with no external dependencies.
Optional: when WebSearch / WebFetch are available, the skill may trace a specific rule's origin (not required).

Input interface

Required: design goal — reaction type plus material system or performance need.
Optional: the structured literature output of catalyst-search (the literature_matrix.md table, supporting conclusions, and validation suggestions).
- With literature — give targeted advice using its specific systems and metrics, and cite the support.
- Without literature — give general advice from the built-in methodology (see references/design_methodology.md).

Methodology system (layered, multi-source, traceable, evolvable)

The methodology continuously fuses multi-source evidence, organized in four layers, and evolves dynamically with new literature.

Full layered knowledge base: references/design_methodology.md (each entry carries a source tag, confidence, and update date).
Dynamic update mechanism: references/methodology_update_protocol.md.
Traceability registry: references/methodology_registry.md (trace by ID to source / DOI).

Four-layer framework

L1 — Theory: structure–property relations, d-band center, Sabatier principle, mechanism typing (AEM / LOM; Volmer–Heyrovsky), activity–stability trade-off, defects as activity, bifunctional synergy.
L2 — Methods and tools: synthesis library (SEA, low-melting doping, ligand anchoring, single-source pyrolysis, Joule heating, templating, grain-boundary engineering), characterization and computation (XRD, STEM, XPS, EXAFS, DFT, ML), structural design (size, ordering, core–shell, support, single- and dual-atom sites).
L3 — Parameters: annealing temperature (450–1150 °C), cooling rate (slow cooling ≤ 2 °C/min), atmosphere, particle size, doping level, composition ratio, post-treatment.
L4 — Validation: RDE → MEA tiered evaluation, applicable standards (TCASMES 400-2024, T/CRES 0030-2025), per-system benchmarks, ordering / stability / scale-up validation.

Source tags (traceability)

[CS] catalyst-search retrieval (with DOI) · [AI] user-acquired via AI / skill · [EXT] external channel (conference, patent, standard, internal data) · [EXP] empirical guess (to verify). Confidence: ★–★★★.

When generating advice, surface the source tag of each cited entry and distinguish "literature-confirmed" from "empirical guess".

Workflow

Parse input. Extract the design goal and collect multi-source evidence — catalyst-search matrix [CS], user-provided [AI], external [EXT]. If the input is incomplete (missing reaction type or material system), ask the user to clarify first; do not guess.
Match rules by layer. Map the goal to L1 theory (mechanism direction) → L2 methods (synthesis / characterization) → L3 parameters (windows) → L4 validation (evaluation path).
Generate advice. Output composition, structure, synthesis, conditions, and validation. Each citation must note its source tag (and DOI) and confidence, and distinguish confirmed from guessed.
Dynamic backfill (evolve, opt-in). Only when the user explicitly asks to update the knowledge base (e.g., "update your methodology" / "remember this"): validate new rules per methodology_update_protocol.md and backfill the skill's own references/design_methodology.md and references/methodology_registry.md (source, DOI, confidence, status). Otherwise keep the built-in methodology read-only and do not modify any skill files. For under-evidenced (only [EXP]) directions, proactively suggest a catalyst-search re-check to upgrade to [CS].
Output. Write the design proposal per templates/design_proposal.md and cite in GB/T 7714 (see catalyst-search's templates/citation_gb7714.md or this skill's own copy).

Output interface

Design proposal covering composition, structure, synthesis, conditions, and validation; format per templates/design_proposal.md.
When citing literature, give a GB/T 7714 citation and label each item "literature-supported" or "empirical guess".

Division of labor

Literature search is catalyst-search's job; this skill does not re-search, only consumes that output or uses its built-in knowledge.
To trace a specific rule's origin, WebSearch / WebFetch may be used (optional).

Notes

Advice must be grounded in the built-in methodology or supplied literature; no ungrounded claims.
Clearly distinguish "literature-confirmed" from "empirical guess".
Advice must include an actionable validation path (characterization, testing) for user verification.
Safety and permissions: by default this skill only reads its own references/ and templates/; it runs no system commands and makes no network requests beyond optional WebSearch / WebFetch; it collects or exfiltrates no user data. Optional self-update: step 4 ("Dynamic backfill") can write to the skill's own references/design_methodology.md and references/methodology_registry.md — but only when you explicitly enable it (e.g., "update your methodology"). It never writes to your own files or project directories.

Example

Input: design an acidic OER catalyst with η₁₀ < 200 mV and stability > 500 h.
Output:
- Composition: Ru@IrOₓ core–shell (high-activity Ru core, dissolution-resistant IrOₓ shell). [CS] ★★★
- Structure: core–shell interfacial charge redistribution, optimizing oxygen-intermediate adsorption on Ru. [CS] ★★★
- Synthesis: Adams fusion for the IrOₓ shell, followed by reductive Ru deposition. [CS] ★★
- Conditions: anneal at 400–500 °C (too high drives Ru into the shell, too low gives insufficient ordering). [EXP] ★
- Validation: RDE for η₁₀ and Tafel slope; ICP-MS for Ru / Ir dissolution monitoring; 30 000-cycle ADT. [CS] ★★★

Project home

GitHub: https://github.com/ANDYPENG09/catalyst-design-skill — source, updates, and issue tracker.
Companion skill: https://github.com/ANDYPENG09/catalyst-search-skill — catalysis literature search.

催化剂设计指导

何时使用

用户咨询:催化剂怎么设计、选什么组分、怎么优化性能、合成路线建议、催化剂配方、电催化剂设计、合成条件怎么选等。
典型输入:设计目标(反应类型 + 材料体系 / 性能诉求)。可选:catalyst-search 的文献检索结果。
覆盖反应:HER、OER、ORR、PEMWE、全解水、光催化、塑料升级回收等。

何时不使用

纯文献检索任务 —— 用 catalyst-search,本技能不重复检索。
非催化领域的材料设计(如电池正极、半导体器件)—— 不适用。
仅需查一个具体 DOI 或论文全文 —— 直接用 WebSearch / WebFetch。

前置条件

本技能自带知识库(references/)与模板(templates/),无需外部依赖即可运行。
可选:当 WebSearch / WebFetch 可用时,可补查特定规律的出处(非必需)。

输入接口

必填:设计目标,含反应类型 + 材料体系或性能诉求。
可选:catalyst-search 的结构化文献输出,即 literature_matrix.md 格式的文献矩阵表 + 支撑结论 + 验证建议。
- 提供了文献 —— 结合文献中的具体体系与指标,给出针对性建议并标注文献支撑。
- 未提供文献 —— 基于内置设计方法论(见 references/design_methodology.md)给出通用建议。

设计方法论体系(分层、多来源、可追溯、可演进)

方法论持续融合多来源证据,按四层体系组织,并可随新增文献动态更新。

完整分层知识库:references/design_methodology.md(每条带来源标签 + 置信度 + 更新日期)。
动态更新机制:references/methodology_update_protocol.md。
可追溯条目台账:references/methodology_registry.md(编号 → 出处 / DOI 逐级溯源)。

四层体系

L1 基础理论层:构效关系、d 带中心、Sabatier 原理、反应机理分型(AEM / LOM;Volmer–Heyrovsky)、活性–稳定性权衡、缺陷即活性、双功能协同。
L2 方法工具层:合成方法库(SEA、低熔点掺杂、配体锚定、单源热解、Joule heating、模板、晶界工程等);表征与计算工具(XRD、STEM、XPS、EXAFS、DFT、ML);结构设计手段(尺寸、有序化、核壳、载体、单/双原子)。
L3 技术参数层:退火温度(450–1150 °C)、降温速率(慢冷 ≤ 2 °C/min)、气氛、粒径、掺杂量、组分配比、后处理。
L4 实践验证层:RDE → MEA 分级评测、相关标准(TCASMES 400-2024、T/CRES 0030-2025)、各体系性能基准、有序度 / 稳定性 / 放大验证。

来源标签(可追溯性)

[CS] catalyst-search 检索文献(含 DOI) · [AI] 用户经 AI / skill 获取 · [EXT] 外部渠道(会议、专利、标准、内部数据) · [EXP] 经验推测(待验证)。置信度:★–★★★。

生成建议时须带出所引条目的来源标签,区分"文献已证实"与"经验推测"。

工作流

解析输入。提取设计目标(反应类型 / 体系 / 性能),并收集多来源证据——catalyst-search 文献矩阵 [CS]、用户经 AI / skill 提供的资料 [AI]、外部渠道 [EXT]。输入不完整(缺反应类型或材料体系)时,先向用户澄清,不臆测。
按四层匹配规律。目标映射到 L1 理论(选机理方向)→ L2 方法工具(选合成 / 表征手段)→ L3 参数(定参数窗口)→ L4 验证(定评测路径)。
生成建议。输出组分选择 + 结构设计 + 合成策略 + 条件优化 + 验证建议。每条引用注明来源标签(及 DOI)与置信度,区分"文献已证实"与"经验推测"。
动态回填(演进,需显式启用)。仅当用户明确要求更新知识库时(例如"更新你的方法论 / 记住这条"),才按 references/methodology_update_protocol.md 校验后回填本技能自身的 references/design_methodology.md 与 references/methodology_registry.md(含来源、DOI、置信、状态)。否则保持内置方法论只读,不修改任何技能文件。证据不足(仅 [EXP])的方向,主动建议调用 catalyst-search 补检以升级为 [CS]。
输出。按 templates/design_proposal.md 输出设计建议书,引用采用 GB/T 7714(格式见 catalyst-search 的 templates/citation_gb7714.md,本技能自带同名模板亦可)。

输出接口

设计建议书(组分 / 结构 / 合成 / 条件 / 验证),格式见 templates/design_proposal.md。
若引用文献,给出 GB/T 7714 引用,并标明"文献支撑"或"经验推测"。

与其他能力分工

文献检索由 catalyst-search 负责;本技能不重复检索,仅消费其输出或基于内置知识给建议。
如需补查特定设计规律的出处,可用 WebSearch / WebFetch(可选,非必需)。

注意事项

建议须有据(内置方法论或所提供文献),不凭空断言。
明确区分"文献已证实"与"经验推测"。
设计建议须附带可执行的验证路径(表征 / 测试),便于用户复核。
安全与权限:默认情况下本技能仅读取自带 references/、templates/;不执行系统命令,除可选的 WebSearch / WebFetch 外不发起网络请求,不收集或外传用户数据。可选的自更新:工作流第 4 步"动态回填"可写入本技能自身的 references/design_methodology.md 与 references/methodology_registry.md 以演进知识库——但仅在您显式启用时(例如"更新你的方法论")才会发生,且绝不会写入您自己的文件或项目目录。

示例

输入:设计一个酸性 OER 催化剂,要求 η₁₀ < 200 mV、稳定 > 500 h。
输出:
- 组分:Ru@IrOₓ 核壳(Ru 核活性高、IrOₓ 壳抗溶)。[CS] ★★★
- 结构:核壳界面电荷重分布,优化 Ru 的含氧中间体吸附。[CS] ★★★
- 合成:Adams Fusion 制 IrOₓ 壳 + 还原沉积 Ru 核。[CS] ★★
- 条件:退火 400–500 °C(过高致 Ru 溶入壳,过低有序度不足)。[EXP] ★
- 验证:RDE 测 η₁₀ / Tafel;ICP-MS 监测 Ru / Ir 溶解;30k 圈 ADT。[CS] ★★★

项目主页

GitHub:https://github.com/ANDYPENG09/catalyst-design-skill —— 源码、更新与 Issue 反馈。
配套技能:https://github.com/ANDYPENG09/catalyst-search-skill —— 催化文献检索。

ANDYPENG09/catalyst-design-skill

catalyst-design-skill とは？

お気に入りのAIに質問する

ドキュメント