← 返回 Skills 市场
OpenClaw Multi-LLM Adapter
作者
michealxie001
· GitHub ↗
· v1.0.0
· MIT-0
109
总下载
0
收藏
0
当前安装
1
版本数
在 OpenClaw 中安装
/install oc-multi-llm
功能描述
Universal adapter for multiple LLM providers. Unified interface for OpenAI, Anthropic, Google Gemini, Ollama, and 100+ providers via LiteLLM. Automatic fallb...
安全使用建议
Key takeaways and actions before installing:
- Expectation mismatch: The documentation promises many providers (LiteLLM, Google Gemini, etc.), but the included code only implements OpenAI, Anthropic, and Ollama. If you need the broader provider support, do not assume it's present — inspect the provider implementations before use.
- Secret handling: The tool expects API keys (OPENAI_API_KEY, ANTHROPIC_API_KEY, GEMINI_API_KEY, OLLAMA_HOST). The package metadata did not declare these required env vars — ensure you know which secrets you'll provide and avoid placing them in shared config files without encryption.
- Exec/read/write permissions: The SKILL.md requests agent tools read/write/exec. Exec lets the agent run shell commands; only enable this if you trust the skill and run it in a sandbox. Prefer running the CLI manually rather than giving the agent autonomous exec rights.
- Dependency installation: requirements.txt will pull packages from PyPI (openai, anthropic, requests). Install dependencies in an isolated virtualenv/container and review package versions.
- Audit and test: Review lib/client.py and scripts/main.py locally to confirm no hidden network endpoints or unexpected behavior. If you want higher assurance, run the skill in an isolated environment (container or VM), limit network access, and avoid giving broad host-level permissions.
- If you plan to use this skill in production or with sensitive data, ask the provider/maintainer for a complete implementation list, explicit metadata declaring required env vars, and justification for the agent exec permission. If those answers are not satisfactory, treat it as untrusted and run locally in a restricted sandbox.
功能分析
Type: OpenClaw Skill
Name: oc-multi-llm
Version: 1.0.0
The skill bundle provides a legitimate universal adapter for multiple LLM providers (OpenAI, Anthropic, Ollama) with support for fallback and model comparison. The code in lib/client.py and scripts/main.py follows standard practices for API integration, including environment variable handling for secrets and structured message processing, with no evidence of malicious intent, data exfiltration, or prompt injection.
能力评估
Purpose & Capability
The README/SKILL.md claims support for many providers (OpenAI, Anthropic, Google Gemini, Ollama, '100+ providers' via LiteLLM). The code (lib/client.py and scripts/main.py) only implements explicit provider classes and wiring for openai, anthropic, and ollama. Declared features (LiteLLM, many provider adapters) are not implemented in the provided files — that's an inconsistency between stated purpose and delivered capability.
Instruction Scope
SKILL.md runtime header lists tools: [read, write, exec]. The instructions show reading a config file (config.yaml) and running CLI commands (python3 scripts/main.py), which explains read/exec usage, but granting exec (arbitrary command execution) to an agent is a broad permission. The SKILL.md and code reference reading arbitrary JSON/YAML config files (tools.json, config.yaml) which could contain API keys or endpoints; those file reads are normal for a CLI but combined with exec permission expands risk if the agent is allowed to run arbitrary shell commands.
Install Mechanism
There is no install spec (instruction-only install), which minimizes automatic remote code download. A requirements.txt is included (openai, anthropic, requests); installing those packages would pull third-party code from PyPI. No direct downloads or obscure URLs are present. This is relatively low risk but installing dependencies means pulling external packages that should be reviewed/installed in a controlled environment.
Credentials
Registry metadata lists no required env vars or primary credential, but SKILL.md and the code expect environment variables such as OPENAI_API_KEY, ANTHROPIC_API_KEY, GEMINI_API_KEY, OLLAMA_HOST, and also allow api_key placeholders in config.yaml. The metadata omission is a mismatch: the skill will look for and use API keys but doesn't declare them up front. Requesting or using multiple provider API keys is reasonable for this tool, but the skill should explicitly declare them. Also the agent-level tools (read/write/exec) could be used to exfiltrate those secrets if combined with network-capable code elsewhere.
Persistence & Privilege
always:false (not force-included) and disable-model-invocation:false (normal) — nothing requests permanent elevated presence. The skill does not modify other skills or system-wide settings in the provided code. However, the SKILL.md's declared exec permission combined with agent autonomy increases blast radius if the agent is permitted to run arbitrary commands; this is a contextual risk rather than a metadata privilege misconfiguration.
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 oc-multi-llm - 安装完成后，直接呼叫该 Skill 的名称或使用
/oc-multi-llm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
v1.0.0
Universal adapter for multiple LLM providers. OpenAI, Anthropic, Gemini, Ollama with fallback support.
元数据
常见问题
OpenClaw Multi-LLM Adapter 是什么？
Universal adapter for multiple LLM providers. Unified interface for OpenAI, Anthropic, Google Gemini, Ollama, and 100+ providers via LiteLLM. Automatic fallb... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 109 次。
如何安装 OpenClaw Multi-LLM Adapter？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install oc-multi-llm」即可一键安装，无需额外配置。
OpenClaw Multi-LLM Adapter 是免费的吗？
是的，OpenClaw Multi-LLM Adapter 完全免费，采用 MIT-0 许可证，可自由下载、安装和使用。
OpenClaw Multi-LLM Adapter 支持哪些平台？
OpenClaw Multi-LLM Adapter 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 OpenClaw Multi-LLM Adapter？
由 michealxie001（@michealxie001）开发并维护，当前版本 v1.0.0。
推荐 Skills