Back to Details

customize

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

Customize

自定义

Use this skill when the user wants to choose a preferred stack for the current project instead of manually copying many external skills.
当用户希望为当前项目选择偏好的技能栈，而非手动复制大量外部技能时，可使用本技能。

Goals

目标

customize
should:
  1. Read the curated registry in 
    skillpacks/skill_dictionary.yaml
    .
  2. Recommend packs and subsets by capability, not just by repo name.
  3. Configure supervision policy alongside skill selection.
  4. Write the result to 
    .co-researcher/skills.yaml
    .
  5. Keep the result easy to read and edit manually.
If 
.co-researcher/skills.yaml
is absent, create it. If present, preserve unrelated fields where possible.
Use 
templates/skills.yaml.template
as the starting shape when you need to create the file from scratch.
When expanding a preset into the full file, merge: preset's 
packs
and 
supervision.mode
/
risk_profile
take precedence; all other fields (
preferences
, full 
supervision
subtree) are filled from template defaults unless the user explicitly changed them.
customize
技能需要：
  1. 读取
    skillpacks/skill_dictionary.yaml
    中的精选注册表。
  2. 基于能力而非仅仓库名称推荐技能包（packs）和子集。
  3. 在选择技能的同时配置监控策略。
  4. 将结果写入
    .co-researcher/skills.yaml
    。
  5. 确保结果便于人工阅读和编辑。
若
.co-researcher/skills.yaml
不存在，则创建该文件；若已存在，则尽可能保留无关字段。
当需要从头创建文件时，以
templates/skills.yaml.template
为基础框架。
将预设扩展为完整文件时，遵循合并规则：预设的
packs
和
supervision.mode
/
risk_profile
优先级最高；所有其他字段（
preferences
、完整
supervision
子树）将使用模板默认值填充，除非用户明确修改。

Inputs

输入

Registry sources

注册表来源

Read:
  • skillpacks/skill_dictionary.yaml
  • skillpacks/presets/*.yaml
These are the source of truth for available packs, curated subsets, and preset bundles.
读取以下文件：
  • skillpacks/skill_dictionary.yaml
  • skillpacks/presets/*.yaml
这些是可用技能包、精选子集和预设包的权威来源。

Existing project state

现有项目状态

If present, also read:
  • .co-researcher/skills.yaml
  • RESEARCH.md
Use them to infer whether the project is literature-heavy, experiment-heavy, publication-focused, or still exploratory.
若存在，还需读取：
  • .co-researcher/skills.yaml
  • RESEARCH.md
通过这些文件推断项目是侧重文献研究、实验、成果发表，还是仍处于探索阶段。

Interaction model

交互模型

Use a short preset-first flow. Ask all questions together when possible.
采用简短的预设优先流程。尽可能一次性提出所有问题。

Required questions

必填问题

  1. Workflow type
    • core-only
    • low-dependency
    • literature-heavy
    • experiment-heavy
    • academic-rigor
    • balanced
    • custom
  2. Dependency tolerance
    • minimal
    • balanced
    • best-in-class even if heavier
  3. Autonomy style
    • manual
    • checkpointed
    • autonomous
    • wild
  4. Resource policy
    • local only
    • APIs allowed
    • remote compute allowed
  5. Preference knobs
    • minimize overlap?
    • prefer subset installs?
    • allow experimental packs?
  1. 工作流类型
    • core-only
    • low-dependency
    • literature-heavy
    • experiment-heavy
    • academic-rigor
    • balanced
    • custom
  2. 依赖容忍度
    • minimal（最低）
    • balanced（平衡）
    • best-in-class even if heavier（即使更复杂也要最优）
  3. 自主风格
    • manual（手动）
    • checkpointed（ checkpoint模式）
    • autonomous（自主）
    • wild（完全自由）
  4. 资源策略
    • local only（仅本地）
    • APIs allowed（允许调用API）
    • remote compute allowed（允许远程计算）
  5. 偏好设置
    • 是否最小化技能重叠？
    • 是否偏好子集安装？
    • 是否允许实验性技能包？

Existing config editing

现有配置编辑

If 
.co-researcher/skills.yaml
already exists, begin by asking:
  • keep current selection and only adjust supervision?
  • adjust packs only?
  • adjust both?
若
.co-researcher/skills.yaml
已存在，首先询问：
  • 保留当前技能选择，仅调整监控设置？
  • 仅调整技能包？
  • 同时调整两者？

Recommendation rules

推荐规则

Capability-first mapping

能力优先映射

  • If the user wants stronger literature search/synthesis, prioritize 
    aris/research-lit
    , 
    aris/research-refine
    , 
    feynman/alpha-research
    , or 
    academic-research-skills/deep-research
    .
  • If the user wants stronger experiment planning, prioritize 
    aris/experiment-plan
    .
  • If the user wants claim/interpretation checking, prioritize 
    aris/result-to-claim
    .
  • If the user wants academic rigor in writing/review, prioritize 
    academic-paper-reviewer
    .
  • If the user wants compute-heavy execution, surface NanoResearch as a niche/heavy option rather than a default.
  • 若用户需要更强的文献搜索/合成能力，优先推荐
    aris/research-lit
    、
    aris/research-refine
    、
    feynman/alpha-research
    或
    academic-research-skills/deep-research
    。
  • 若用户需要更强的实验规划能力，优先推荐
    aris/experiment-plan
    。
  • 若用户需要验证结论/解读，优先推荐
    aris/result-to-claim
    。
  • 若用户需要写作/审稿的学术严谨性，优先推荐
    academic-paper-reviewer
    。
  • 若用户需要计算密集型执行，将NanoResearch作为小众/重型选项推荐，而非默认选项。

Dependency warnings

依赖警告

  • feynman/alpha-research
    requires AlphaXiv authentication (
    feynman alpha login
    ). Always surface this when recommending 
    literature-heavy
    or any feynman skill.
  • nanoresearch/project-experiment
    requires local GPU or cluster access. Always surface this when recommending 
    experiment-heavy
    with NanoResearch.
  • feynman/alpha-research
    需要AlphaXiv认证（执行
    feynman alpha login
    ）。推荐
    literature-heavy
    工作流或任何feynman技能时，必须提示此信息。
  • nanoresearch/project-experiment
    需要本地GPU或集群访问权限。推荐
    experiment-heavy
    工作流搭配NanoResearch时，必须提示此信息。

Default posture

默认策略

  • Always keep 
    core
    enabled.
  • Prefer curated subsets over full-pack installs.
  • Avoid recommending heavy overlapping packs unless the user explicitly wants them.
  • 始终启用
    core
    技能包。
  • 优先选择精选子集而非完整技能包安装。
  • 除非用户明确要求，否则避免推荐高度重叠的技能包。

Preset choice

预设选择

Prefer a preset when the user's answers fit one cleanly. Fall back to custom pack selection only when necessary.
balanced
should map to the 
balanced
preset in 
skillpacks/presets/balanced.yaml
.
当用户的答案完全匹配某一预设时，优先选择该预设。仅在必要时才回退到自定义技能包选择。
balanced
工作流应映射到
skillpacks/presets/balanced.yaml
中的
balanced
预设。

Output file

输出文件

Write or update:
text
.co-researcher/skills.yaml
Expected shape:
yaml
version: 1
profile: balanced

selection:
  preset: balanced

enabled_packs:
  core:
    enabled: true
  aris:
    enabled: true
    mode: subset
    selected_skills:
      - research-lit
      - research-refine
      - experiment-plan
      - result-to-claim

preferences:
  minimize_overlap: true
  prefer_low_dependency: true
  prefer_subset_installs: true
  allow_experimental_packs: false

supervision:
  mode: checkpointed
  risk_profile: balanced
  approval_policy:
    install_new_skills: ask
    update_registry: ask
    modify_config: ask
    launch_experiments: ask
    use_external_api: ask_first_use
    run_long_review_loops: ask
    create_commits: ask
  resource_policy:
    local_only: false
    cloud_allowed: false
    cluster_allowed: false
  budget_policy:
    enabled: true
    api_budget_level: medium
  execution_limits:
    max_review_rounds: 2
    require_confirmation_for_new_pack_imports: true
写入或更新：
text
.co-researcher/skills.yaml
预期格式：
yaml
version: 1
profile: balanced

selection:
  preset: balanced

enabled_packs:
  core:
    enabled: true
  aris:
    enabled: true
    mode: subset
    selected_skills:
      - research-lit
      - research-refine
      - experiment-plan
      - result-to-claim

preferences:
  minimize_overlap: true
  prefer_low_dependency: true
  prefer_subset_installs: true
  allow_experimental_packs: false

supervision:
  mode: checkpointed
  risk_profile: balanced
  approval_policy:
    install_new_skills: ask
    update_registry: ask
    modify_config: ask
    launch_experiments: ask
    use_external_api: ask_first_use
    run_long_review_loops: ask
    create_commits: ask
  resource_policy:
    local_only: false
    cloud_allowed: false
    cluster_allowed: false
  budget_policy:
    enabled: true
    api_budget_level: medium
  execution_limits:
    max_review_rounds: 2
    require_confirmation_for_new_pack_imports: true

Required behavior after recommendation

推荐后的必填操作

Before writing the file:
  1. Show the recommended preset or custom stack.
  2. Show selected packs and selected skills.
  3. Show supervision mode and any notable approval/resource settings.
  4. Explain tradeoffs briefly.
  5. Ask for confirmation.
After writing the file:
  1. Tell the user where the config was written.
  2. Summarize the chosen stack.
  3. Mention that the file is intended to be hand-editable.
写入文件前：
  1. 展示推荐的预设或自定义技能栈。
  2. 展示选中的技能包和技能。
  3. 展示监控模式及所有重要的审批/资源设置。
  4. 简要说明权衡利弊。
  5. 请求用户确认。
写入文件后：
  1. 告知用户配置文件的写入位置。
  2. 总结所选技能栈。
  3. 提及该文件支持人工编辑。

Constraints

约束条件

  • Do not install external skills silently.
  • Do not mutate 
    skillpacks/skill_dictionary.yaml
    ; that belongs to 
    evolve
    .
  • Do not recommend full-pack imports unless the user clearly asks for them.
  • Prefer simple, project-local configuration over hidden session-only state.
  • 不得静默安装外部技能。
  • 不得修改
    skillpacks/skill_dictionary.yaml
    ；该文件属于
    evolve
    技能的管理范围。
  • 除非用户明确要求，否则不推荐完整技能包导入。
  • 优先选择简单的项目本地配置，而非隐藏的仅会话状态。