Health-Mate

Health-Mate is an executable OpenClaw health-management skill for bilingual daily wellness tracking. It is not an instruction-only prompt bundle. The package includes Python scripts and shell runners that read structured markdown health logs, generate daily and weekly PDF reports, and optionally send report payloads to configured webhook endpoints.

Purpose

• Read structured markdown health logs from MEMORY_DIR
• Parse meals, hydration, exercise, steps, symptoms, and custom monitoring sections
• Generate localized daily and weekly health reports
• Render local PDF files
• Optionally deliver final report payloads to DingTalk, Feishu, or Telegram

Installation

Install the Python dependencies locally:

pip install -r requirements.txt

Main runtime libraries:

• reportlab
• pillow
• matplotlib

Required Configuration

Required

• MEMORY_DIR This skill requires an explicit MEMORY_DIR environment variable. If MEMORY_DIR is missing, the Python entrypoints and shell runners now exit instead of falling back to any default path.

Optional

• TAVILY_API_KEY Enables extra search context for next-day planning only.

• DINGTALK_WEBHOOK

• FEISHU_WEBHOOK
• TELEGRAM_BOT_TOKEN

• TELEGRAM_CHAT_ID These are used only when outbound delivery is intentionally enabled.

• REPORT_WEB_DIR

• REPORT_BASE_URL Used only when PDFs should be copied to a public directory and exposed with a downloadable URL.

• ALLOW_RUNTIME_FONT_DOWNLOAD Disabled by default. Only enable this if you explicitly want the PDF renderer to download NotoSansSC-VF.ttf when no bundled or system CJK font is available.

Local File-System Behavior

The skill writes local files during normal operation. This is expected behavior.

• Reads markdown logs from MEMORY_DIR
• Reads config/.env only when the bundled shell runners are used and that file exists
• Writes generated PDFs to the local reports/ directory unless a different public copy target is configured
• Writes runtime logs to the local logs/ directory
• May create a temporary English memory mirror in the system temp directory when Chinese memory is rendered without a bundled local assets/NotoSansSC-VF.ttf
• May place the fallback font file in the local assets/ directory only when ALLOW_RUNTIME_FONT_DOWNLOAD is explicitly enabled

Runtime Network Behavior

Outbound network activity is limited and conditional:

• Webhook delivery is performed only if DingTalk, Feishu, or Telegram credentials are configured
• Tavily requests are performed only if TAVILY_API_KEY is configured
• Fallback font download from raw.githubusercontent.com occurs only if ALLOW_RUNTIME_FONT_DOWNLOAD=true is explicitly set

If you do not want any runtime font download, leave ALLOW_RUNTIME_FONT_DOWNLOAD unset and pre-populate assets/NotoSansSC-VF.ttf if you need guaranteed Chinese PDF output. If the local bundled font is missing, Chinese-source reports fall back to English output with an explicit rendering notice.

AI Runtime Notes

• If openclaw is available in the runtime environment, the skill can request AI commentary, next-day planning, and weekly review text
• If openclaw is not available, the skill falls back to deterministic local logic so report generation still completes
• When multiple conditions are configured, both the LLM path and the fallback path evaluate the combined condition set
• Weekly output can summarize additional configured or dynamically discovered monitoring sections, such as medication or biochemistry

Command Intent

• /health Generate the localized daily report for today or for a specified date

• /health summary Generate the localized weekly review anchored to a specified date

Agent Behavior

When this skill is active, the assistant may be natural and helpful in chat, but it must be strict and mechanical when writing files under MEMORY_DIR.

• Never write encouragement, analysis, commentary, summaries, or conversational filler into the memory file
• Use one language per record block
• Chinese and English are both supported, but field labels must not be mixed inside the same block
• If time or calories are uncertain, estimate conservatively while preserving the exact structure

Memory Write Protocol

When writing into MEMORY_DIR, the model must behave like a mechanical data recorder. Advice, commentary, and conversational guidance belong in chat only and must never be written into the memory file. The following structure must be preserved exactly because downstream parsing depends on fixed regex patterns.

If an LLM is used to write local health memories, the same rules should be mirrored into soul.md or the runtime system prompt so the runtime layer and the skill layer do not drift apart.

Non-Negotiable Rules

1. Meals, hydration blocks, and exercise blocks must and may only use level-3 headings with time markers: ### Label (around HH:MM) or ### 标签（约 HH:MM）
2. Food lines must use the exact arrow format: - Item portion → approx. XXXkcal
3. Hydration blocks must contain exactly two lines and nothing else: - Water intake: XXXml - Cumulative: XXXml/targetml
4. Single exercise records must use an exercise-specific level-3 heading such as ### Afternoon Cycling (around 17:17) or ### 下午骑行（约 17:17）, then list distance, duration, and burn only
5. Step tracking must use one level-2 heading only: ## Today Steps followed by: - Total steps: XXXX steps
6. Extra modules such as medication are allowed, but they must stay under level-2 headings and contain raw bullet data only
7. File output must never include evaluation words such as Assessment, Status, Summary, 评估, 状态, 总结, or any emoji
8. The memory file must never contain chat-style comments, LLM explanations, motivational language, or extra fields that are not part of the template
9. Chinese and English templates are both valid, but the model must not mix them in the same record block

Chinese Template

# 2026-03-20 健康记录

### 早餐（约 08:30）
- 燕麦片 50g → 约 190kcal
- 脱脂牛奶 250ml → 约 87kcal

### 上午（约 09:45）
- 饮水量：300ml
- 累计：300ml/2000ml

### 下午骑行（约 17:17）
- 距离：10公里
- 耗时：47分钟
- 消耗：约 300kcal

## 今日步数
- 总步数：8500 步

English Template

# 2026-03-20 Health Log

### Breakfast (around 08:30)
- Oatmeal 50g → approx. 190kcal
- Skim milk 250ml → approx. 87kcal

### Morning (around 09:45)
- Water intake: 300ml
- Cumulative: 300ml/2000ml

### Afternoon Cycling (around 17:17)
- Distance: 10km
- Duration: 47min
- Burn: approx. 300kcal

## Today Steps
- Total steps: 8500 steps

Bilingual Parsing Rules

• Daily and weekly reports may be generated from either the Chinese template or the English template
• File headings and field labels are parsed bilingually
• Internal code and config prefer canonical English values
• Chinese display output remains fully supported when language is set to zh-CN

Operational Review Notes

For security and deployment review, the following should be considered expected behavior rather than hidden behavior:

• local PDF and log file creation
• optional loading of config/.env by the bundled shell runners
• optional webhook delivery
• optional Tavily search usage
• optional opt-in fallback font download when ALLOW_RUNTIME_FONT_DOWNLOAD is enabled

Changelog

v1.3.5 — 2026-03-21

• Added automatic English PDF fallback when assets/NotoSansSC-VF.ttf is missing for Chinese-source reports, using a temporary exported English memory mirror and an explicit rendering notice
• Kept runtime font download opt-in only, with documentation and metadata aligned to the actual installation and runtime behavior
• Updated release metadata, config examples, and documentation versioning for the 1.3.5 release

v1.3.0 — 2026-03-20

• Added a shared bilingual language layer for prompts, parsing, PDF rendering, and shell delivery
• Added bilingual markdown parsing for meals, hydration, exercise, steps, symptoms, and custom sections
• Hardened the Memory Write Protocol with Chinese and English templates and stronger anti-commentary rules
• Updated metadata, config examples, and documentation to align with the bilingual release