← 返回 Skills 市场
hubentu

Coala Client

作者 qhu · GitHub ↗ · v0.1.2
cross-platform ⚠ suspicious
623
总下载
0
收藏
0
当前安装
1
版本数
在 OpenClaw 中安装
/install coala
功能描述
How to use the coala-client CLI for chat with LLMs, MCP servers, and skills. Use when the user asks how to use coala, run coala chat, add MCP servers, import...
使用说明 (SKILL.md)

Coala Client

Part of the coala ecosystem. CLI for chat with OpenAI-compatible LLMs (OpenAI, Gemini, Ollama) and MCP (Model Context Protocol) servers. Supports importing CWL toolsets as MCP servers, importing skills.

Config paths

  • MCP config and toolsets: ~/.config/coala/mcps/
    • mcp_servers.json — server definitions
    • \x3Ctoolset>/ — per-toolset dirs with run_mcp.py and CWL files
  • Skills: ~/.config/coala/skills/ (one subfolder per imported source)
  • Env: ~/.config/coala/env (optional; key=value for providers and MCP env)

Quick start

  1. Init (first time)
    coala init — creates ~/.config/coala/mcps/mcp_servers.json and env.

  2. Set API key
    e.g. export OPENAI_API_KEY=... or export GEMINI_API_KEY=.... Ollama needs no key.

  3. Chat
    coala or coala chat — interactive chat with MCP tools.
    coala ask "question" — single prompt with MCP.

  4. Options
    -p, --provider (openai|gemini|ollama|custom), -m, --model, --no-mcp.

MCP: CWL toolsets

No API key needed for MCP import, list, or call — only for chat/ask with an LLM.

  • Import (creates toolset under ~/.config/coala/mcps/\x3CTOOLSET>/ and registers server):
    coala mcp-import \x3CTOOLSET> \x3CSOURCES...> or alias coala mcp ...
    SOURCES: local .cwl files, a .zip, or http(s) URLs to a .cwl or .zip.
    Requires the coala package where the MCP server runs (for run_mcp.py).

  • List
    coala mcp-list — list server names.
    coala mcp-list \x3CSERVER_NAME> — print each tool’s schema (name, description, inputSchema).

  • Call
    coala mcp-call \x3CSERVER>.\x3CTOOL> --args '\x3CJSON>'
    Example: coala mcp-call gene-variant.ncbi_datasets_gene --args '{"data": [{"gene": "TP53", "taxon": "human"}]}'

Skills

  • Import (into ~/.config/coala/skills/, one subfolder per source):
    coala skill \x3CSOURCES...>
    SOURCES: GitHub tree URL (e.g. https://github.com/owner/repo/tree/main/skills), zip URL, or local zip/dir.

  • In chat
    /skill — list installed skills.
    /skill \x3Cname> — load skill from ~/.config/coala/skills/\x3Cname>/ (e.g. SKILL.md) into context.

Chat commands

  • /help, /exit, /quit, /clear
  • /tools — list MCP tools
  • /servers — list connected MCP servers
  • /skill — list skills; /skill \x3Cname> — load a skill
  • /model — show model info
  • /switch \x3Cprovider> — switch provider

MCP on/off

  • All off: coala --no-mcp (or coala ask "..." --no-mcp).
  • One server off: remove its entry from ~/.config/coala/mcps/mcp_servers.json.
  • On: default when --no-mcp is not used; add or restore servers in mcp_servers.json.

Providers and env

Set provider via -p or env PROVIDER. Set keys and URLs per provider (e.g. OPENAI_API_KEY, GEMINI_API_KEY, OLLAMA_BASE_URL). Optional: put vars in ~/.config/coala/env.
coala config — print current config paths and provider/model info.

安全使用建议
This skill appears to be a legitimate helper for the coala-client CLI, but be cautious about importing toolsets or skills from remote URLs or zipped archives: those artifacts can contain code (run_mcp.py or other scripts) that the client may execute when you start an MCP server or load a skill. Before installing or using: 1) verify the origin of the 'uv' package (where 'coala-client' comes from), 2) avoid importing skills/toolsets from untrusted/unverified URLs, 3) inspect downloaded zips/local files before import, 4) prefer running MCP/toolsets in an isolated environment or sandbox, and 5) only provide LLM API keys to providers you trust and consider using --no-mcp when you do not want external tools to run.
功能分析
Type: OpenClaw Skill Name: coala Version: 0.1.2 The `SKILL.md` file describes the `coala-client` tool's ability to import MCP toolsets and skills from arbitrary HTTP(S) URLs. Specifically, `coala mcp-import` and `coala skill` commands allow fetching `.cwl` files, `.zip` archives, or skill directories from remote locations. This introduces a significant supply chain risk and potential remote code execution (RCE) vulnerability, as the AI agent could be prompted to download and process untrusted remote content, even though the skill itself does not demonstrate malicious intent.
能力评估
Purpose & Capability
Name/description (coala-client CLI for chat, MCP, and skills) match the declared binary requirement (coala-client) and the documented actions (init, chat, mcp-import, skill import). No unrelated credentials or unrelated binaries are requested.
Instruction Scope
Instructions are specific to coala-client usage and reference config paths under ~/.config/coala/. However, the SKILL.md explicitly allows importing CWL/toolsets and skills from arbitrary http(s) URLs or zip files and describes running MCP servers (run_mcp.py). That implies downloading and potentially executing third‑party code from remote sources — behavior coherent with the tool but risky if sources are untrusted.
Install Mechanism
Install spec uses a 'uv' package named 'coala-client' that creates the coala-client binary. The manifest does not show a raw URL download or archive extraction, but the origin/resolver for the 'uv' package is not described here — verify the package registry/source before installing.
Credentials
No required environment variables or credentials are declared. The documentation sensibly notes optional provider keys (OPENAI_API_KEY, GEMINI_API_KEY, OLLAMA_BASE_URL) needed only for LLM provider access; these are proportional and expected.
Persistence & Privilege
Skill is not always-enabled and does not request persistent elevated privileges or modification of other skills. It uses per-user config paths under ~/.config/coala/, which is appropriate for a CLI tool of this type.
如何使用
  1. 确保已安装 OpenClaw(本地或 Docker 部署)
  2. 在对话框中输入安装命令:/install coala
  3. 安装完成后,直接呼叫该 Skill 的名称或使用 /coala 触发
  4. 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
v0.1.2
- Removed mentions of the optional sandbox run_command tool. - Updated description and feature lists to reflect removal of sandbox-related content. - All references to sandbox functionality have been omitted for clearer documentation.
元数据
Slug coala
版本 0.1.2
许可证
累计安装 0
当前安装数 0
历史版本数 1
常见问题

Coala Client 是什么?

How to use the coala-client CLI for chat with LLMs, MCP servers, and skills. Use when the user asks how to use coala, run coala chat, add MCP servers, import... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 623 次。

如何安装 Coala Client?

在 OpenClaw 或 Claude Code 对话框中运行命令「/install coala」即可一键安装,无需额外配置。

Coala Client 是免费的吗?

是的,Coala Client 完全免费(开源免费),可自由下载、安装和使用。

Coala Client 支持哪些平台?

Coala Client 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 Coala Client?

由 qhu(@hubentu)开发并维护,当前版本 v0.1.2。

💬 留言讨论