Oneskills — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited Oneskills (Plugin) and scored it 100/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 0 high-severity and 0 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 0 flagged
Every scanned point with the score it earned and what moved between them.
Score rose 4 points between these scans.
The primary manifest — the file an agent reads to learn what this artifact does.
<p align="center"> A skills library for AI-native scientific research built around <strong>OneScience</strong>. </p>
<p align="center"> Reusable skills for <strong>workflow orchestration</strong>, <strong>coding</strong>, <strong>runtime submission</strong>, and <strong>environment setup</strong>. </p>
<p align="center"> Works with <strong>Claude Code</strong>, <strong>Codex CLI</strong>, <strong>Trae</strong>, and other skill-based agents through optional integration adapters. </p>
OneSkills 是面向 AI4S(AI for Science)场景的通用技能仓库。
它把科研开发中的工作流理解、角色协作、代码生成、远程安装与运行经验,整理成可复用的 SKILL.md 能力模块,供不同智能体消费。
仓库核心是:
skills/:通用技能references/:通用参考资料integrations/:可选的智能体适配层catalog/manifest.json:官方 skill 清单(供 OneCode Catalog ingest).well-known/skills/index.json:远程 skill 索引(供 skills.urls 拉取)公开版本信息:
VERSION:当前发布版本号skills/VERSION:技能包版本号RELEASE_NOTES.md:本次对外发布说明推荐主链:
onescience-workflow -> onescience-role -> onescience-skill -> onescience-coder -> onescience-runtime
-> onescience-installer其中:
onescience-workflow:理解用户真实科研任务onescience-role:做角色协作拆分onescience-skill:选择最小执行链onescience-coder:负责代码实现与工程改造onescience-runtime:负责统一运行闭环,内部包含 discover / preflight / execute / diagnoseonescience-installer:负责环境安装与修复,内部包含 discover / precheck / install / verify统一入口约定:
onescience-workflow,先询问具体科研目标,不直接拉起完整执行链。onescience-coder、onescience-runtime、onescience-installer 时,优先使用对应 skill 完成工作。当前公开架构中,运行链路由 onescience-runtime 统一处理,并按 3 层理解:
execution_mode / access_mode / execution_channellocal、ssh_slurm 还是 scnet_mcpbackend_idremote_slurm 链路下,根据 hardware_profile 与 selector 命中具体 backendruntime_profile_ref / install_profile_ref当前关键边界:
onescience-runtime 是唯一公开运行入口onescience-installer 是唯一公开安装/修复入口onescience-runtime 内部固定为 discover -> preflight -> execute -> diagnoseonescience-installer 内部固定为 discover -> precheck -> install -> verifyinstall_profile_ref 不等于 installer backend 名称当前稳定 runtime backend:
slurm_dcuslurm_gpuslurm_gpu_multinode_torchrunslurm_cpu当前公开远程通道:
ssh_slurm:remote_slurm + sshscnet_mcp:remote_direct + cloud_api统一入口约定:
onescience-workflow,先询问具体科研目标,不直接拉起完整执行链。onescience-coder、onescience-runtime、onescience-installer 时,优先使用对应 skill 完成工作。onescience-workflowonescience-roleonescience-skillonescience-coderonescience-runtimeonescience-installer当前 oneskills 可以直接支持以下用户需求:
onescience-runtime / onescience-installer 内部完成环境识别、预检与约束归一化onescience-runtime 处理;ssh_slurm 链路基于 onescience.json、模板和 backend/profile registry 确定运行 backend 与 profile,scnet_mcp 链路通过本地已接入的 SCnet MCP 上传脚本、提交任务、轮询状态并下载日志;若本地未安装该服务,则 scnet_mcp 通道不可用onescience-runtime 的 diagnose 阶段识别模型测试、Earth DataPipe 测试或完整训练 / 推理流程测试路径OneScience更直观地说,用户可以直接提出这类问题:
推荐使用统一安装器:
python3 install/install_oneskills.py --agent codex --project /your/project支持的目标包括:
codexclaudetraeopencodegeneric(需额外提供 --skills-dir)常用方式:
python3 install/install_oneskills.py --agent codex --project /your/project
python3 install/install_oneskills.py --agent claude --project /your/project
python3 install/install_oneskills.py --agent trae --project /your/project
python3 install/install_oneskills.py --agent opencode --project /your/projectOpenCode 默认会把 skill 安装到 .opencode/skills/,与 OpenCode 自动发现路径一致;通常不必修改 opencode.jsonc。若使用了非默认 --namespace-root,可参考 .opencode/opencode.jsonc.snippet 合并 skills.paths。详见 integrations/opencode/README.md。
开发期如果希望技能目录始终跟随当前仓库更新,可使用软链接模式:
python3 install/install_oneskills.py --agent codex --project /your/project --mode symlink如果还要把运行资产一并装到项目根目录:
python3 install/install_oneskills.py --agent codex --project /your/project --profile runtime默认是 basic 档位,只安装 skills / references / integrations。 如果项目要实际提交远程任务,改用 --profile runtime。 如果目标是 codex,安装器还会自动补一层 ~/.codex/skills/onescience-* bridge,让安装后默认可被当前 Codex 发现。
卸载:
python3 install/install_oneskills.py --agent codex --project /your/project --uninstall更多参数见:
install/README.mdinstall/install_oneskills.pydocs/user-guides/claude_codex_plugin_install.md跨平台建议:
macOS / Linux:可使用 copy 或 symlinkWindows:建议使用默认 copy;若显式传入 --mode symlink,安装器会自动降级为 copy常见问题见:
install/README.md如果你不想使用安装器,也可以手动复制。
git clone https://github.com/onescience-ai/oneskills.git
mkdir -p /your/project/.codex/oneskills/skills
mkdir -p /your/project/.codex/oneskills/references
mkdir -p /your/project/.codex/oneskills/integrations/codex
cp -r oneskills/skills/* /your/project/.codex/oneskills/skills/
cp -r oneskills/references/* /your/project/.codex/oneskills/references/
cp oneskills/integrations/generic-agent.md /your/project/.codex/oneskills/integrations/generic-agent.md
cp oneskills/integrations/codex/README.md /your/project/.codex/oneskills/integrations/codex/README.md如果你是手动安装到 codex,还需要自己补一层 ~/.codex/skills/onescience-* bridge;因此更建议直接使用安装器。
推荐使用 Claude Code 插件安装:
/plugin marketplace add https://github.com/onescience-ai/oneskills
/plugin install oneskills@oneskills安装后重启 Claude Code,或在支持的版本中执行 /reload-plugins。
本地开发调试时,也可以 clone 后把仓库根目录作为 marketplace:
git clone https://github.com/onescience-ai/oneskills.git/plugin marketplace add ./oneskills
/plugin install oneskills@oneskillsgit clone https://github.com/onescience-ai/oneskills.git
mkdir -p /your/project/.trae/oneskills/skills
mkdir -p /your/project/.trae/oneskills/references
cp -r oneskills/skills/* /your/project/.trae/oneskills/skills/
cp -r oneskills/references/* /your/project/.trae/oneskills/references/你也可以使用 generic 模式安装到任意技能目录:
python3 install/install_oneskills.py --agent generic --project /your/project --skills-dir .agent/oneskills/skills如果你想在不破坏当前产品分层的前提下扩展自定义技能,建议先阅读:
docs/user-guides/extend_domain_experience.mddocs/open-source/custom_skill_contribution.md推荐原则:
integrations/generic-agent.mdintegrations/codex/README.mdintegrations/claude/integrations/opencode/README.md.claude-plugin/:Claude Code 插件元数据与 marketplace 示例.codex/INSTALL.md:Codex 原生 skills 发现安装方式Claude Code 推荐执行 /plugin marketplace add https://github.com/onescience-ai/oneskills。本地测试时执行 /plugin marketplace add ./oneskills,不要添加 ./oneskills/.claude-plugin。这样 marketplace 中的 source: "./" 才会指向仓库根目录,并安装完整的 skills/ 目录。
本仓库公开通用技能、公开参考资料与用户可用文档。
如果你想了解如何在本仓库里扩展自定义技能,优先阅读:
docs/open-source/custom_skill_contribution.mdLICENSECONTRIBUTING.mdSECURITY.md~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.