技能介绍
先看中文介绍;官方 description 原文单独保留,不改写 SKILL.md。
做什么
互动CLI用于Uni-Mol分子属性预测培训和推论工作流程.
何时用
官方 description 未单独写出 Use when。按规范,代理会在用户任务与这段 description 的关键词匹配时激活本技能。
代理如何加载
按 Agent Skills 渐进披露:启动时只加载 name 与 description(约 100 token);任务匹配后才读入整份 SKILL.md 正文;scripts/、references/、assets/ 仅在需要时再读。 本文件正文结构:Uni-Mol Tools - Molecular Property Prediction CLI、Description、Key Features、Common Commands、Project Management、Create a new project。
文件分析
文件分析:这是一份仅含 SKILL.md 的指令型技能,代理激活后整份正文进入上下文。
官方 description(原文)
Interactive CLI for Uni-Mol molecular property prediction training and inference workflows.
Uni-Mol Tools - Molecular Property Prediction CLI Description Key Features Common Commands Project Management Create a new project List all projects Switch to a project Training Train a classification model Train a regression model Model Management
来源分类:skills.sh agent-skill
来源与网址
namecli-anything-unimol-tools
descriptionInteractive CLI for Uni-Mol molecular property prediction training and inference workflows.
发现技能 客户端向 Agent 提供名称与描述目录。
匹配与调用 用户指定或任务匹配后,载入 SKILL.md 指令。
按需加载 按步骤读取参考文档、使用脚本与素材。
具体调用语法与可用工具以目标 Agent 客户端为准。 查看调用机制说明 ↗
先选择目标 Agent 和安装范围,保留技能包的附属文件,安装后检查客户端能否发现该技能。
交给 Agent 安装
复制安装指令给支持 Agent Skills 的代理,确认其中的目标目录与客户端匹配。
把 Agent Skill「cli-anything-unimol-tools」安装到我的项目:SKILL.md 原文与官方 description 见 https://zicq.com/zh/skills/skl-cbc94a101b55d5c4-Cli-%E4%BB%BB%E6%84%8F-Unimol-%E5%B7%A5%E5%85%B7.html
请存为 .cursor/skills/cli-anything-unimol-tools/SKILL.md 或 .claude/skills/cli-anything-unimol-tools/SKILL.md,frontmatter 的 name 与 description 保持原样,不要改写。
复制安装提示词 GitHub 完整包 ↗
终端安装 · Skills CLI
需要 Node.js 与 npx。先查看仓库技能列表,确认实际名称。
npx skills add 'https://github.com/hkuds/cli-anything' --list
复制查看命令
目标客户端
安装时交互选择 Codex Claude Code Cursor
安装范围
当前项目 我的所有项目
npx skills add 'https://github.com/hkuds/cli-anything' --skill 'cli-anything-unimol-tools' 复制安装命令
CLI 会交互选择目标 Agent,默认安装到项目;用户级安装使用 -g。先通过查看命令核对仓库内容,再用 npx skills list 检查已安装技能。
SKILL.md 原文 可复制,不会改动源文件
复制 SKILL.md
恢复原文
阅读排版
---
name: "cli-anything-unimol-tools"
description: >-
Interactive CLI for Uni-Mol molecular property prediction training and inference workflows.
---
# Uni-Mol Tools - Molecular Property Prediction CLI
**Package**: `cli-anything-unimol-tools`
**Command**: `python3 -m cli_anything.unimol_tools`
## Description
Interactive CLI for training and inference of molecular property prediction models using Uni-Mol Tools. Supports 5 task types: binary classification, regression, multiclass, multilabel classification, and multilabel regression.
## Key Features
- **Project Management**: Organize experiments with named projects
- **5 Task Types**: Classification, regression, multiclass, multilabel variants
- **Model Tracking**: Automatic performance history and rankings
- **Smart Storage**: Analyze usage and clean up underperformers
- **JSON API**: Full automation support with `--json` flag
## Common Commands
### Project Management
```bash
# Create a new project
project create --name drug_discovery
# List all projects
project list
# Switch to a project
project switch --name drug_discovery
```
### Training
```bash
# Train a classification model
train --data-path train.csv --target-col active --task-type classification --epochs 10
# Train a regression model
train --data-path train.csv --target-col affinity --task-type regression --epochs 10
```
### Model Management
```bash
# List all trained models
models list
# Show model details and performance
models show --model-id
# Rank models by performance
models rank
```
### Storage & Cleanup
```bash
# Analyze storage usage
storage analyze
# Automatic cleanup of poor performers
cleanup auto
# Manual cleanup with criteria
cleanup manual --max-models 10 --min-score 0.7
```
### Prediction
```bash
# Make predictions with a trained model
predict --model-id --data-path test.csv
```
## Data Format
CSV files must contain:
- `SMILES` column: Molecular structures in SMILES format
- Target column(s): Values to predict (name specified via `--target-col`)
Example:
```csv
SMILES,target
CCO,1
CCCO,0
CC(C)O,1
```
## Task Types
1. **classification**: Binary classification (0/1)
2. **regression**: Continuous value prediction
3. **multiclass**: Multiple class classification
4. **multilabel_classification**: Multiple binary labels
5. **multilabel_regression**: Multiple continuous values
## JSON Mode
Add `--json` flag to any command for machine-readable output:
```bash
python3 -m cli_anything.unimol_tools --json models list
```
Output format:
```json
{
"status": "success",
"data": [...],
"message": "..."
}
```
## Interactive Mode
Launch without commands for interactive REPL:
```bash
python3 -m cli_anything.unimol_tools
```
Features:
- Tab completion
- Command history
- Contextual help
- Project state persistence
## Test Data
Example datasets available at:
https://github.com/545487677/CLI-Anything-unimol-tools/tree/main/unimol_tools/examples
Includes data for all 5 task types.
## Requirements
- Python 3.8+
- PyTorch 1.12+
- Uni-Mol Tools backend
- 4GB+ RAM (8GB+ recommended for training)
## Installation
```bash
cd unimol_tools/agent-harness
pip install -e .
```
## Documentation
- **SOP**: [UNIMOL_TOOLS.md](../UNIMOL_TOOLS.md)
- **Quick Start**: [docs/guides/02-QUICK-START.md](../docs/guides/02-QUICK-START.md)
- **Full Documentation**: [docs/README.md](../docs/README.md)
## Testing
```bash
cd docs/test
bash run_tests.sh --unit -v # Unit tests (67 tests)
bash run_tests.sh --full -v # Full test suite
```
## Performance Tips
- Start with 10 epochs for initial experiments
- Use smaller batch sizes if memory is limited
- Monitor storage with `storage analyze`
- Use `models rank` to identify best performers
- Clean up regularly with `cleanup auto`
## Troubleshooting
- **CUDA errors**: Reduce batch size or use CPU mode
- **CSV not recognized**: Verify SMILES column exists
- **Low accuracy**: Try more epochs or adjust learning rate
- **Storage full**: Run `cleanup auto` to free space
## Related
- **Uni-Mol Tools**: https://github.com/dptech-corp/Uni-Mol/tree/main/unimol_tools
- **Uni-Mol Paper**: https://arxiv.org/abs/2209.11126
- **CLI-Anything**: https://github.com/HKUDS/CLI-Anything
相关技能
Agent 工作流
创造有效技能指南。 当用户想创造出新的技能(或更新现有的技能),以专业知识,工作流程,或工具集成来扩展克洛德的能力时,应该使用这种技能.
Agent 工作流
使用ClawdHub CLI搜索,安装,更新并发布从taladhub.com的代理技能. 需要获取苍蝇上的新技能时使用,将安装的技能同步到最新版本或特定版本,或者发布 npm-instainddhub CLI 的新/更新的技能文件夹.
Agent 工作流
管弦乐团多代理团队,任务设定周期,交接协议,审查工作流程. 使用时间: (1)建立2+特派员队伍,具有不同专业,(2)确定任务路线和生命周期(收录框_ spec_建设_审查_完成),(3)在特派员之间制定交接协议,(4)建立审查和质量关口,(5)管理特派员之间的交流和文物共享.
Agent 工作流
Spec-first,TDD,子代理驱动的软件开发工作流程. 当:(1)构建任何新功能或应用——触发脑暴_计划_子代理执行回路,(2)调试出一个bug或测试失败——触发系统性的根起过程,(3)用户说"让我们构建","帮助我计划","我想添加X",或"这个被打破",(4)完成一个功…