← 返回 Skills 市场
103
总下载
0
收藏
1
当前安装
6
版本数
在 OpenClaw 中安装
/install ai-model-team
功能描述
AI Model Team - Multi-model prediction system (Kronos + Chronos-2 + TimesFM + VADER) for OKX crypto and US stocks. Features error handling, retry logic, HTTP...
使用说明 (SKILL.md)
AI Model Team
Multi-model collaborative prediction system: Kronos + Chronos-2 + TimesFM + VADER FinBERT
Models
- Kronos-base: K-line specialist (NeoQuasar)
- Chronos-2: Macro cycle prediction (Amazon)
- TimesFM-2.5-200M: General time series (Google)
- VADER FinBERT: Financial sentiment (\x3C1s load)
Supported Assets
| Type | Examples | Data Source |
|---|---|---|
| Crypto | BTC, ETH, SOL | OKX API |
| US Stocks | NVDA, AAPL, MSFT | Yahoo Finance |
Usage
# 4-model analysis
python scripts/model_team.py BTC-USDT-SWAP --models kronos,chronos-2,timesfm,finbert
# Run tests
python -m pytest tests/ -v
Robustness
- Full error handling (Kronos, Chronos, TimesFM)
- Reddit API retry (3 retries, 429 handling)
- HTTP caching (lru_cache, 60s TTL)
- Yahoo Finance for US stocks
- Unit tests (pytest, 9 tests)
安全使用建议
This package is plausible for model-based trading, but there are mismatches you should resolve before installing or running it with real credentials: 1) The registry metadata declares no required credentials, yet the code expects environment variables and may call OKX (or run execution code). Assume it will try to read a local .env and other skill dirs. 2) Inspect scripts/execution.py, scripts/security.py, and scripts/post_install.py to confirm whether the skill performs live trades or can trigger withdrawals; do not set ALLOW_WITHDRAWAL=true. 3) Run it first in an isolated environment (separate VM or container) and create a fresh virtualenv as instructed; use read-only or simulated API keys (no funds) if you must provide credentials. 4) Search for networking endpoints and webhook targets (ALERT_WEBHOOK) and verify they are expected. 5) Prefer to request explicit metadata from the developer: which env vars are required (OKX API key/secret, any webhook URLs), what install scripts will modify, and whether any autonomy/execution will run automatically. If you cannot verify those, treat the skill as untrusted for use with real accounts or secrets.
功能分析
Type: OpenClaw Skill
Name: ai-model-team
Version: 2.9.2
The skill bundle provides a sophisticated trading prediction system but employs several high-risk execution patterns. Specifically, `scripts/post_install.py` performs a monkey patch by writing code directly into the `__init__.py` file of the installed `timesfm` library to resolve dependency conflicts. Additionally, `scripts/timesfm_adapter.py` and `scripts/run_team.py` use `subprocess.run` to execute multi-line Python scripts constructed as strings, and `scripts/okx_data_provider.py` invokes an external `okx` CLI tool. While these behaviors appear aligned with the stated purpose of managing complex ML dependencies and fetching market data, the use of library modification and dynamic code execution represents a significant attack surface.
能力标签
能力评估
Purpose & Capability
The skill's stated purpose is trading prediction and (per README/CHANGELOG) it includes data providers for OKX and execution/risk-control code. A trading/execution-capable package would normally require declared credentials (API keys) and explicit install/permission steps. The registry metadata lists no required env vars or credentials, which is inconsistent with the presence of execution.py, security.py, okx_data_provider.py and README instructions referencing OKX and other OKX-related skills.
Instruction Scope
SKILL.md and README focus on running the Python scripts and tests, but the code loads a .env file (config.load_env_file()), reads environment variables for many operational controls (e.g., ALERT_WEBHOOK, ALLOW_WITHDRAWAL, API-path defaults), and references other skill directories via AI_HEDGE_PATH. Reading a local .env and other skill paths can surface secrets/config outside the skill's manifest; the instructions do not call this out or require explicit approval.
Install Mechanism
There is no formal install spec in the registry (instruction-only), but the README requires creating a .venv, installing pinned dependencies (including git+https installs from GitHub commit hashes) and places a patched timesfm copy under .venv/src/timesfm. GitHub installs by commit are traceable (moderate risk). The README references a scripts/install.sh which is not listed in the manifest — that mismatch and the need to mutate a .venv/src directory raise operational friction and risk if users run undocumented scripts.
Credentials
The skill manifest declares no required env vars, yet config.py and other modules read many environment variables (ensemble weights, ALLOW_WITHDRAWAL, ALERT_WEBHOOK, REQUEST_TIMEOUT_SEC, AI_HEDGE_PATH, etc.). Trading/OKX interaction implies API keys (not declared). The skill therefore expects access to sensitive secrets without declaring them — a proportionality and transparency problem. Also ALLOW_WITHDRAWAL and other execution-related flags exist and could enable sensitive actions if set.
Persistence & Privilege
always:false (good). However the project contains post_install.py and instructions that persist patched code into a .venv/src directory and references other skill directories (AI_HEDGE_PATH). Those behaviors imply file writes and lasting local modifications if the user follows the README. The registry metadata does not make these persistence actions explicit.
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 ai-model-team - 安装完成后，直接呼叫该 Skill 的名称或使用
/ai-model-team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
v2.9.2
v2.9.2: skill directory structure fix + v2.9.1 docs complete
v2.9.1
v2.9.1: OKX pagination + Chronos dtype fix + TimesFM persistence + full docs
v2.9.0
v2.9.0: Fix CL crude oil data source - Yahoo Finance CL was Colgate stock, now uses OKX. Document Kronos insufficient data policy.
v2.8.0
v2.8.0: Fix ETH price bug
v2.7.0
v2.7.0: Retry publish with .clawhubignore
v2.6.0
v2.6.0
元数据
常见问题
AI Model Team 是什么？
AI Model Team - Multi-model prediction system (Kronos + Chronos-2 + TimesFM + VADER) for OKX crypto and US stocks. Features error handling, retry logic, HTTP... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 103 次。
如何安装 AI Model Team？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install ai-model-team」即可一键安装，无需额外配置。
AI Model Team 是免费的吗？
是的，AI Model Team 完全免费，采用 MIT-0 许可证，可自由下载、安装和使用。
AI Model Team 支持哪些平台？
AI Model Team 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 AI Model Team？
由 Yirong（@erongcao）开发并维护，当前版本 v2.9.2。
推荐 Skills