← 返回 Skills 市场
355
总下载
0
收藏
1
当前安装
3
版本数
在 OpenClaw 中安装
/install soma
功能描述
Expert guide for participating in the SOMA network — a decentralized system that trains a foundation model through competition. Provides data submission work...
安全使用建议
This skill appears to be a comprehensive SOMA contributor guide and does require signing keys, HuggingFace access, and S3 credentials to operate — that is expected for the described workflows. However: 1) the registry metadata does not list these env vars even though the SKILL.md requires them; ask the publisher to fix the manifest so you know exactly what will be needed. 2) Do NOT use mainnet/private production keys for testing — follow the skill's suggestion to use testnet keys and explicitly confirm which network a given key targets. 3) Be cautious about exporting wallet secret keys and pushing .env contents to any cloud provider: prefer using short-lived, least-privilege credentials scoped to a single S3 bucket, and rotate keys after testing. 4) Understand that submission data and even encrypted weights may be published with public-read ACLs per protocol — do not submit proprietary or regulated data. 5) If you plan to run any recommended scripts (federated submitter, commit/reveal), review the quickstart repo code (the docs reference GitHub) before running, and ensure Modal or any orchestration service you use is trusted and configured securely. If the author cannot explain why the registry metadata omits required env vars or refuses to update it, treat the omission as a red flag and do not provide secrets.
功能分析
Type: OpenClaw Skill
Name: soma
Version: 1.1.0
The skill bundle facilitates participation in the 'SOMA' decentralized AI network, requiring the agent to install a CLI via a high-risk 'curl | bash' command (sup.soma.org) and manage several sensitive credentials including signing keys, HuggingFace tokens, and S3 access keys. While the documentation includes security best practices like using testnet keys and scoped permissions, the reliance on an unverified external installer combined with the collection of high-value secrets presents a significant risk of credential harvesting or supply chain compromise.
能力评估
Purpose & Capability
The SKILL.md and references explicitly describe on-chain signing, S3 uploads, and HuggingFace dataset access — so requesting SOMA_SECRET_KEY, HF_TOKEN, and S3 credentials is coherent with the skill's stated purpose. However, the skill registry metadata lists no required environment variables or binaries even though the runtime docs require the soma CLI and several env vars; that metadata/manifest mismatch is an inconsistency that should be corrected.
Instruction Scope
The instructions stay within SOMA-specific workflows (data submission, scoring, commit-reveal, uploading weights). They do instruct potentially risky actions: exporting secret keys (wallet export), storing secrets in a local .env and pushing them to a cloud secret store (Modal), and uploading submission data / encrypted weights with public-read ACLs. Those actions are functionally consistent with the protocol but increase exposure of sensitive material and could leak private data if misconfigured.
Install Mechanism
This is an instruction-only skill with no install spec or code files to execute — lowest installer risk. The docs recommend installing CLI via an external installer (sup) and Python packages via pip, which is typical and expected; nothing in the bundle performs arbitrary downloads or writes to disk at install time.
Credentials
The SKILL.md expects multiple highly sensitive environment variables (SOMA_SECRET_KEY, HF_TOKEN, S3_ACCESS_KEY_ID, S3_SECRET_ACCESS_KEY) which are proportionate to the described capabilities (on-chain signing, dataset access, artifact upload). The concern is the manifest/registry does not declare these required env vars, and the runtime guidance encourages exporting secret keys and pushing them to a remote secret store — both increase risk if a user accidentally supplies mainnet credentials or misconfigures public ACLs.
Persistence & Privilege
always:false and disable-model-invocation:false (normal). The skill is instruction-only and does not request permanent presence or attempt to modify other skills or system settings. There is no install step that gives it elevated persistence or system-wide privileges.
如何使用
- 确保已安装 OpenClaw（本地或 Docker 部署）
- 在对话框中输入安装命令：
/install soma - 安装完成后，直接呼叫该 Skill 的名称或使用
/soma触发 - 根据 Skill 的参数说明提供必要输入，即可获得结构化输出
版本历史
v1.1.0
Declare required env vars and security notes, add merge_coins docs, deploy+trigger flow
v0.1.1
- Updated "Getting Started" section for a streamlined, step-by-step workflow focusing on fast deployment of the data submitter without needing local GPU or localnet.
- Added a credentials table with clear explanations of each required setting, why it's needed, and where to obtain it.
- Clarified that data submission is now the recommended entry point; emphasized how to deploy directly to Modal for GPU compute.
- Expanded the quick decision tree, offering a more direct route for new contributors and clarifying next actions.
- Improved instructions on how to claim rewards and what to do after initial setup, increasing accessibility for new users.
v0.1.0
Initial release of the SOMA skill — expert guidance for participating in the SOMA decentralized model network.
- Provides detailed workflows for data submission, model training, and claiming rewards.
- Includes guidance on SDK code generation, environment setup, and CLI usage.
- Offers strategic advice for maximizing SOMA rewards and choosing competitive niches.
- Links to official documentation and repository for further reference.
- Restricts scope to SOMA network-specific tasks; not for general ML or unrelated PyTorch/JAX questions.
元数据
常见问题
Soma 是什么？
Expert guide for participating in the SOMA network — a decentralized system that trains a foundation model through competition. Provides data submission work... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 355 次。
如何安装 Soma？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install soma」即可一键安装，无需额外配置。
Soma 是免费的吗？
是的，Soma 完全免费，采用 MIT-0 许可证，可自由下载、安装和使用。
Soma 支持哪些平台？
Soma 跨平台运行，可在任意部署了 OpenClaw / Claude Code 的环境中使用（cross-platform）。
谁开发了 Soma？
由 Cfuaqz（@cfuaqz）开发并维护，当前版本 v1.1.0。
推荐 Skills