什么是 喝水教练？

喝水教练是为 Openclaw Skills 生态系统构建的专业级健康扩展程序。它通过结构化记录、基于体重的个性化目标设定和主动式指导，赋能 AI 智能体管理用户的水分平衡。通过利用自然语言处理技术，该技能允许用户维护准确的饮水和身体指标记录，而无需手动输入的繁琐操作。

此技能专为不仅仅满足于简单记录的用户设计；它提供了一个习惯养成的逻辑框架。它直接与智能体的记忆集成，保持健康数据的持久性和私密性，确保您的 AI 助手成为您健康旅程中信息充沛的伙伴。无论您是为了健身目标还是医疗需求而追踪饮水量，喝水教练都能提供精确健康监测所需的技术基础设施。

下载入口:https://github.com/openclaw/skills/tree/main/skills/oristides/water-coach

安装与下载

1. ClawHub CLI

从源直接安装技能的最快方式。

2. 手动安装

将技能文件夹复制到以下位置之一

优先级：工作区 > 本地 > 内置

3. 提示词安装

将此提示词复制到 OpenClaw 即可自动安装。

喝水教练 应用场景

• 使用自然语言描述记录饮水量，例如“我刚刚喝了两杯水”。
• 根据体重和用户偏好自动计算每日补水目标。
• 设置自定义通知时间表，在白天的特定时间段接收提醒。
• 追踪历史身体指标，包括体重、身高和体脂率。
• 生成详细的每周和每月表现报告，以分析补水趋势。

1. AI 智能体解析用户输入，并将随意的描述转换为精确的度量单位 (ml)。
2. 智能体执行命令行界面 (CLI)，将条目记录到位于工作空间记忆中的 CSV 审计追踪文件中。
3. 系统利用心跳功能每 30 分钟进行一次动态阈值检查，识别用户是否落后于目标进度。
4. 定时任务 (cron jobs) 触发综合分析简报，总结不同时段（早晨、午餐、下午、晚上）的表现。
5. 用户可以通过审计与报告摄入量对话相关的特定消息 ID 来核实任何条目。

喝水教练 配置指南

要开始使用这款 Openclaw Skills 补水工具，请按照以下步骤操作：

# 初始化设置流程
water_coach.py water setup

# 配置您的体重以设定默认目标
water_coach.py water set_body_weight 80 --update-goal

# 验证当前设置状态
water_coach.py water status

请确保您的环境中已启用 python3 和心跳功能。

喝水教练 数据架构与分类体系

喝水教练在技能逻辑和用户数据之间保持了清晰的分离。所有数据都存储在智能体工作空间记忆目录中。

name: water-coach
description: "Hydration tracking and coaching skill. Use when user wants to track water intake, get reminders to drink water, log body metrics, or get analytics on hydration habits."
compatibility: "Requires python3, openclaw cron feature, heartbeat feature"
metadata: {"clawdbot":{"emoji":"??"} 
  author: oristides

?? Water Coach

First-time Setup

Follow the first setup here references/setup.md

?? How Other Agents Should Interact

First-time Setup Check

water setup

? Don't Ask

• Reminder schedules after first setup (user already configured)

? Do Ask

• "How much water did you drink?"
• Only weight/goal (first time)

CLI Structure

water_coach.py   [options]

Namespaces: water | body | analytics

Data Format

CSV Format

logged_at,drank_at,date,slot,ml_drank,goal_at_time,message_id

Key Rules:

• drank_at is MANDATORY - always required
• If user doesn't specify drank_at → assume drank_at = logged_at
• Cumulative is calculated at query time (not stored)
• Use drank_at to determine which day counts

Details at references/log_format.md

Audit Trail

Every water log entry captures:

• message_id: Links to the conversation message where user requested the log
• Auto-capture: CLI automatically gets message_id from session transcript
• Proof: Use water audit to get entry + conversation context

# Check proof of a water entry
water audit msg_123
# Returns: entry data + surrounding messages for context

?? Privacy Notice: The audit feature can read your conversation transcripts, but only when you explicitly run water audit . This is off by default (audit_auto_capture: false).

# Edit water_config.json and set:
"audit_auto_capture": true

How it works:

• Water log always saves the message_id (regardless of this setting) ?
• When you run water audit :
  • If false: Shows entry data only (message_id saved, but no context read)
  • If true: Also reads transcript to show conversation context ("User said: I drank 500ml")

Why disable it? If you discuss sensitive topics and don't need proof of intake, leave it off.

Daily Commands

# Water
water status                                      # Current progress (calculated from drank_at)
water log 500                                    # Log intake (drank_at = now)
water log 500 --drank-at=2026-02-18T18:00:00Z  # Log with past time
water log 500 --drank-at=2026-02-18T18:00:00Z --message-id=msg_123
water dynamic                                    # Check if extra notification needed
water threshold                                  # Get expected % for current hour
water set_body_weight 80                        # Update weight + logs to body_metrics
water set_body_weight 80 --update-goal          # + update goal
water audit                         # Get entry + conversation context

# Body
body log --weight=80 --height=1.75 --body-fat=18
body latest          # Get latest metrics
body history 30     # Get history

# Analytics
analytics week       # Weekly briefing (Sunday 8pm)
analytics month     # Monthly briefing (2nd day 8pm)

Rules (MUST FOLLOW)

1. ALWAYS use CLI - never calculate manually
2. LLM interprets first - "eu tomei 2 copos" → 500ml → water log 500
3. Threshold from CLI - run water threshold, don't hardcode
4. GOAL is USER'S CHOICE - weight × 35 is just a DEFAULT suggestion:
   • At setup: Ask weight → suggest goal → CONFIRM with user
   • On weight update: Ask "Want to update your goal to the new suggested amount?"
   • User can set any goal (doctor's orders, preference, etc.)

Config Tree

water-coach/
├── SKILL.md              ← You are here
├── scripts/
│   ├── water_coach.py   ← Unified CLI
│   └── water.py         ← Core functions
├── data/                 ← DO NOT USE - keep skill code separate from user data
└── references/
    ├── setup.md
    ├── dynamic.md
    └── log_format.md

?? IMPORTANT: Data Location

User data is stored in the AGENT WORKSPACE, NOT in the skill folder!

Example path: /home/oriel/.openclaw/workspace/memory/data/

Why? Keeps user data separate from skill code — makes backups, migrations, and skill updates easier.

Notifications Schedule

Notification Rules (MUST FOLLOW)

• USER CONFIGURES THEIR SCHEDULE — Agent should ask user: "What times do you want water reminders?" and respect that
• NO "no log" skip logic — Always send notifications when scheduled, don't skip because user hasn't logged water
• Notifications STIMULATE logging — That's the point! Don't assume user will log on their own

Quick Reference

Dynamic Scheduling details

→ references/dynamic.md

?? Bug Fix (Feb 2026)

The water dynamic command had a bug where the hourly notification counter wouldn't reset when the hour changed. This is now fixed:

• The script now checks if the current hour differs from last_extra_hour and resets the counter accordingly
• This ensures notifications work correctly after hour boundaries (e.g., 4PM)

?? Bug Fix (Feb 2026) - Analytics

The analyticsPM → 5 week and analytics month commands had a bug:

• Was looking for non-existent cumulative_ml column in CSV
• Fixed to sum ml_drank per day instead

? How to Build Good Weekly/Monthly Reports

Use these functions (don't reinvent):

These call get_week_stats() and get_month_stats() in water.py.

When updating analytics functions, follow these rules:

1. Include ALL days, even with 0ml

# In get_week_stats() / get_month_stats()
# Include every day in the range, not just days with data
for i in range(days):
    d = (date.today() - timedelta(days=i)).strftime("%Y-%m-%d")
    ml = by_date.get(d, {}).get("ml", 0)  # Default to 0, not skip

2. Calculate true average

# Average = total_ml / ALL days (including zeros), not just tracked days
avg_ml = total_ml / days  # e.g., 15440ml / 7 days = 2205ml/day

3. Show all days in table format

| Dia | ML | % | Status |
| Sab 22 | 2250ml | 67.7% | ?? |
| Seg 17 | 0ml | 0.0% | ? |

This gives users an accurate picture of their habits!

Tests

python3 -m pytest skills/water-coach/scripts/test/test_water.py -v

Example

User: "eu tomei 2 copos"
Agent: (LLM interprets: 2 copos ≈ 500ml)
Agent: exec("water_coach.py water log 500")
→ Python logs to CSV

Agent Evaluations → evaluation/AGENT.md