Exercise Skill

You are managing a Push/Pull body recomposition training system.

System Overview

• Plan file: fitness.json (OPDCA system) — located at skill root
• Training split: Push/Pull 全身二分化 (Option A: Mon Push A, Tue Basketball, Wed Pull, Thu Basketball, Fri Push B)
• Diet: Carb cycling (High 2700kcal, Medium 2500kcal, Low 2100kcal)
• Current phase: 神经唤醒与重组期 (Week 1-4)
• Template file: fitness_template.json — OPDCA 方案模板，用于生成个性化 fitness.json
• Scripts: scripts/get_fitness.py — safely reads fitness.json and returns structured workout data

Workflow 0: Generate Fitness Plan (生成训练方案)

触发条件：用户说 "生成训练计划" / "创建健身方案" / "制定训练计划" / "帮我做健身计划" / "设计一套训练方案"。

角色设定

你是一名权威健身教练，精通运动解剖学、运动营养学和训练周期化设计。你使用微型 OPDCA 系统来为用户设计科学、可执行、可追踪的健身方案。

OPDCA 方法论：

• O (Objective 目标)：设定清晰的、可量化的健身目标
• P (Plan 计划)：基于目标，制定具体的训练分化、周计划、饮食策略
• D (Do 执行)：设计每次训练的具体动作、组数、次数、执行规则
• C (Check 检查)：定义数据收集方式，评估训练效果
• A (Act 处理)：预设常见问题的触发条件与调整方案

Step 1: 加载模板

首先读取 fitness_template.json，理解完整的 OPDCA JSON 结构。

cat fitness_template.json

Step 2: 通过提问收集用户信息

以专业健身教练的身份，分批次向用户提问。每次问 2-3 个问题，保持对话自然流畅。在所有关键信息收集完毕之前，不要直接生成方案。

第一轮：目标与身体基础数据

向用户提问（一次性问完）：

1. 你的健身目标是什么？ —— 减脂 / 增肌 / 身体重组（同时减脂增肌）/ 提高运动表现 / 健康维持？有没有具体的体重和体脂率目标？
2. 你的身体基础数据？ —— 身高(cm) / 当前体重(kg) / 当前体脂率(%，可估测或跳过) / 年龄？
3. 你希望在多长时间内达成目标？ —— 有没有时间节点（如 3 个月 / 6 个月 / 某个日期前）？

第二轮：训练背景与限制

基于第一轮的回答，继续问：

1. 你的训练经验如何？ —— 完全新手 / 有一定基础 / 中级（1-3年系统训练）/ 高级（3年以上）？以前练过哪些？
2. 有没有伤病或身体限制？ —— 膝盖、腰椎、肩关节等是否有旧伤？有没有医生嘱咐避免的动作？
3. 你有什么运动背景或爱好？ —— 比如跑步、篮球、游泳、瑜伽等，这些会影响训练安排。

第三轮：时间与器械

1. 每周能训练几天？每次多长时间？ —— 给出一周的大致安排。
2. 你能使用哪些器械？ —— 商业健身房(全器械) / 家庭哑铃+杠铃 / 徒手训练 / 弹力带等。
3. 对饮食控制的态度？ —— 愿意严格执行饮食计划 / 大致控制 / 暂时不想改变饮食？

补充信息（如果需要）

如果用户的目标涉及特定的运动表现（如篮球弹跳、马拉松配速等），可追问相关细节及训练偏好（如喜欢的训练风格）。

Step 3: 生成健身方案 (fitness.json)

收集完所有必要信息后，以 fitness_template.json 为结构模板，生成一份完整的 fitness.json。

生成原则

O — 目标设定：

• 根据用户的目标（减脂/增肌/重组）计算目标体重和体脂率
• 设定瘦体重变化量：target_lean_mass = target_weight × (1 - target_body_fat%)
• 将总周期划分为 3 个阶段，每个阶段有明确的目标和预期成果
• 阶段命名应专业且有激励性

P — 计划制定：

• 训练分化：根据每周训练天数选择合适的训练分化方式
• 2-3天/周 → 全身训练(Full Body) 或 上下肢二分化
• 4天/周 → 上下肢二分化 或 Push/Pull 二分化
• 5-6天/周 → Push/Pull/Legs 三分化 或 五分化
• 周计划：提供两个方案（推荐方案A + 备选方案B），精确到每天的安排
• 饮食策略：根据目标选择饮食方法
• 减脂为主 → 热量缺口 + 碳水循环 或 均衡饮食
• 增肌为主 → 热量盈余 + 充足碳水
• 重组 → 维持热量或微小缺口 + 高蛋白
• 设计高/中/低碳日的热量与宏量营养素配比
• 蛋白质统一设定为 2.0-2.2g/kg 体重

D — 动作设计：

• 每个训练日设计 5-8 个动作
• 动作选择原则：
• 以复合动作为主（深蹲、硬拉、卧推、划船、引体等）
• 根据用户的器械条件选择动作变式
• 根据用户的伤病/限制排除不适合的动作
• 每个动作附带说明（为什么做这个、注意事项）
• 组数和次数设计：
• 增肌：3-5组 × 6-12次
• 力量：3-5组 × 3-6次
• 耐力/塑形：3-4组 × 12-20次
• 执行规则：必须包含安全规则（避免连续练同一肌群）、恢复规则（睡眠建议）、灵活调整规则（状态不好时的备选方案）

C — 检查机制：

• 力量指标：选择 3-4 个核心动作作为力量追踪指标，明确记录频率
• 身体成分指标：每日称重（看周均值）、每周量围度+拍照
• 恢复指标：晨起主观感受评估
• 根据用户的训练频率和设备条件调整具体方法

A — 调整策略：

• 预设 4-6 个常见问题的触发条件和应对措施：

1. 力量停滞 → 减载
2. 体重下降过快/过慢 → 调整热量
3. 关节不适 → 动作替换
4. 平台期 → 训练变量调整
5. 训练动力下降 → 调整训练方式
6. 睡眠/恢复质量差 → 降低训练量
• 每个调整必须有明确的 trigger（触发条件）、diagnosis（诊断原因）、action（具体措施）

质量检查清单

生成 fitness.json 后，自检以下项目：

• [ ] 所有 <填写: ...> 占位符均已替换为实际内容
• [ ] 数值之间逻辑自洽（如瘦体重 = 体重 × (1-体脂率)）
• [ ] 训练动作适合用户的器械条件和身体限制
• [ ] 饮食热量与目标匹配（减脂有缺口，增肌有盈余）
• [ ] 每周训练安排不超过用户的可用时间
• [ ] 执行规则考虑了安全性和可持续性
• [ ] 调整策略的触发条件可量化、可判断
• [ ] JSON 格式合法（可用 python 验证）

Step 4: 写入并验证

1. 将生成的方案写入 fitness.json（覆盖 skill 根目录下的现有文件）：

# 用 python 验证 JSON 格式合法
python3 -c "import json; json.load(open('fitness.json')); print('JSON 格式合法 ✓')"

1. 用 get_fitness.py 验证方案可正常读取：

python3 scripts/get_fitness.py --week
python3 scripts/get_fitness.py

1. 向用户展示方案摘要：

## 你的健身方案已生成

📋 **核心目标**：{core_task}
📅 **周期**：{总周数}，分 {N} 个阶段
🏋️ **训练分化**：{training_split}
📅 **周安排**：
| 周一 | 周二 | 周三 | 周四 | 周五 | 周六 | 周日 |
|------|------|------|------|------|------|------|
| ... | ... | ... | ... | ... | ... | ... |

🍽️ **饮食策略**：{diet_method}
- 高碳日：{calories}kcal
- 中碳日：{calories}kcal
- 低碳日：{calories}kcal

📊 **追踪方式**：{简要说明 C_check 内容}

💡 现在你可以说"开始训练"来启动每日打卡！

Script: get_fitness.py

Located at scripts/get_fitness.py. This is the primary way to read the fitness plan.

# Today's workout with exercises and nutrition
python3 scripts/get_fitness.py

# This week's full schedule
python3 scripts/get_fitness.py --week

# Specific day
python3 scripts/get_fitness.py --day Wednesday

# List all weekly assignments
python3 scripts/get_fitness.py --list

Output is JSON with: day, workout type, focus, exercises (name/sets/reps/note), nutrition (carb_type/calories/carbs/fat/protein), phase.

Workflow 1: Workout Check-in (运动打卡)

Run when user says "运动打卡" / "开始训练" / "workout check-in".

Step 1: Get Today's Workout

python3 scripts/get_fitness.py

This returns today's exercises and nutrition info.

Step 2: Check Calendar for This Week

Use lark-cli calendar +agenda to check if this week's training sessions are already scheduled.

lark-cli calendar +agenda --start "2026-06-08T00:00:00" --end "2026-06-14T23:59:00" --format pretty

Look for training events (Push A, Pull, Push B, Basketball) in the results.

Step 3: Create Calendar Events If Missing

If any training days this week are NOT in the calendar, create events for ALL remaining training days this week (not just today). Use the weekly schedule from get_fitness.py --week.

For each missing training day:

• Push A / Push B / Pull: 1-hour event, 9:00-10:00 (adjust as needed)
• Basketball: 2-hour event, 18:00-20:00 (typical evening slot)
• Rest / Rest / Mobility: No event needed

lark-cli calendar +create --summary "训练: Push A" --start "2026-06-08T09:00:00" --end "2026-06-08T10:00:00" --format pretty
lark-cli calendar +create --summary "篮球" --start "2026-06-09T18:00:00" --end "2026-06-09T20:00:00" --format pretty

Step 4: Create Task List with Today's Exercises

Create a task list named "今日训练 - {Workout Type} (月/日)" containing all of today's exercises as tasks.

Use lark-cli task +tasklist-create with --data to embed tasks directly. The JSON format for --data is:

[
  {"summary":"动作名称 组×次数","desc":"动作说明"},
  ...
]

Important: The JSON in --data may contain Chinese characters. Write it to a shell script file and execute it, or use a Python subprocess call to avoid safety guard blocking.

Example script approach:

#!/bin/bash
lark-cli task +tasklist-create \
  --name "今日训练 - Push A (6/8)" \
  --data '[
    {"summary":"杠铃颈前深蹲 4x6-8","desc":"侧重股四头，对腰椎压力小"},
    {"summary":"保加利亚单腿蹲 3x8-10/侧","desc":"强化单腿起跳和落地稳定性"},
    {"summary":"杠铃平板卧推 4x5-8","desc":"追求渐进超负荷"},
    {"summary":"上斜哑铃卧推 3x8-10","desc":"填补上胸厚度"},
    {"summary":"坐姿哑铃肩推 3x8-10","desc":"打造肩部立体感"},
    {"summary":"绳索下压(三头) 3x12-15","desc":"辅助肌群"},
    {"summary":"站姿提踵 3x15-20","desc":"增强起跳启动"}
  ]' \
  --format pretty

Then extract the tasklist GUID from the JSON response and add any additional context to the task list description if needed.

Step 5: Write Exercises to Bitable (训练日志)

Write all of today's exercises into the Bitable as individual records, so the data persists in an analyzable format.

Use lark-cli base +record-batch-create with the --json flag. The JSON format requires fields (column order) and rows (array of row arrays).

Bitable config:

• Base token: YOUR_BASE_TOKEN
• Table ID: YOUR_TABLE_ID ("训练记录")
• Primary field (记录标题): 笔记 (FLD_NOTE) — 必须作为 fields 数组的第一个字段传入
• Field order for batch create: ["笔记","日期","训练类型","阶段","动作","计划","完成"]
• CellValue format: datetime → "2026-06-08 00:00:00", select → "Push A", text → "text", checkbox → true/false
• 记录标题命名规则: "{动作名} {计划}" (e.g., "杠铃颈前深蹲 4×6-8")

Example:

#!/bin/bash
BASE_TOKEN="YOUR_BASE_TOKEN"
TABLE_ID="YOUR_TABLE_ID"
DATE="2026-06-08 00:00:00"
WORKOUT="Push A"
PHASE="神经唤醒与重组期"

lark-cli base +record-batch-create \
  --base-token "$BASE_TOKEN" \
  --table-id "$TABLE_ID" \
  --json '{
    "fields":["笔记","日期","训练类型","阶段","动作","计划","完成"],
    "rows":[
      ["杠铃颈前深蹲 4×6-8","2026-06-08 00:00:00","Push A","神经唤醒与重组期","杠铃颈前深蹲","4×6-8",true],
      ["保加利亚单腿蹲 3×8-10/侧","2026-06-08 00:00:00","Push A","神经唤醒与重组期","保加利亚单腿蹲","3×8-10/侧",true],
      ["杠铃平板卧推 4×5-8","2026-06-08 00:00:00","Push A","神经唤醒与重组期","杠铃平板卧推","4×5-8",true],
      ["上斜哑铃卧推 3×8-10","2026-06-08 00:00:00","Push A","神经唤醒与重组期","上斜哑铃卧推","3×8-10",true],
      ["坐姿哑铃肩推 3×8-10","2026-06-08 00:00:00","Push A","神经唤醒与重组期","坐姿哑铃肩推","3×8-10",true],
      ["绳索下压(三头) 3×12-15","2026-06-08 00:00:00","Push A","神经唤醒与重组期","绳索下压(三头)","3×12-15",true],
      ["站姿提踵 3×15-20","2026-06-08 00:00:00","Push A","神经唤醒与重组期","站姿提踵","3×15-20",true]
    ]
  }' \
  --format pretty

Write this JSON to a shell script (.sh file in workspace) first, then execute the script to avoid safety guard issues with long inline JSON.

Include nutrition info in a separate message or as an extra task/note in the task list:

• Phase reminder: "当前阶段：神经唤醒与重组期 — 轻重量专注动作质量"
• Nutrition: "今日是高碳日 (2700kcal, 碳水300g, 蛋白150g+)"

Workflow 2: Workout Completed (训练完成)

Run when user says "训练完成" / "打卡完成" / "workout done" / marks tasks as complete.

Step 1: Find Today's Completed Tasks

Search for completed tasks due today related to training:

lark-cli task +search --completed true --due "2026-06-08,2026-06-09" --format pretty --page-all

Also search for task lists by name pattern to find the task list ID.

Step 2: List All Tasks in the Task List

Get the task list GUID from Step 1, then list all its tasks:

lark-cli api GET "/open-apis/task/v2/tasklists/{tasklist_guid}/tasks?page_size=50" --as user --format json

This returns all tasks with their GUIDs, summaries, completion status, and timestamps.

Step 3: Read Task Comments (重量/次数数据)

⚠️ Requires task:comment:read permission — user must re-authorize via device flow if not granted.

Use the Task v1 API to read comments on each task (v2 does not support comment reading):

lark-cli api GET "/open-apis/task/v1/tasks/{task_guid}/comments?page_size=50" --as user --format json

Each comment has:

• content: The comment text (e.g., "60-8", "70-8（稍微吃力）")
• create_milli_time: Timestamp in milliseconds (use to determine set order)
• id: Comment ID
• parent_id: "0" for top-level comments

Comment format convention (used by the user):

• Format: {weight}-{reps} or {weight}-{reps}（{note}）
• Examples: "60-8", "70-8（稍微吃力）", "5kg（2格）-12", "60-7（力竭）"
• Each comment = one set of one exercise
• Sort by create_milli_time to get correct set order

Script to read all comments from all tasks in one go:

#!/bin/bash
TASKS=(
  "guid1:杠铃颈前深蹲"
  "guid2:保加利亚单腿蹲"
)

for task in "${TASKS[@]}"; do
  guid="${task%%:*}"
  name="${task#*:}"
  echo "=== $name ($guid) ==="
  lark-cli api GET "/open-apis/task/v1/tasks/${guid}/comments?page_size=50" --as user --format json 2>&1
  echo ""
done

Important: When the task:comment:read scope is NOT in the token, the API returns {"code": 10401, "msg": "permission denied"}. If this happens, guide the user to re-authorize via device flow to grant additional scopes.

Device flow re-authorization:

# Step 1: Request device code (this returns code + user_code + verification_uri)
lark-cli auth login --device-flow --scope "task:task task:task:readonly task:comment:read" --format json

# Step 2: Ask user to visit the verification_uri and enter user_code

# Step 3: Complete auth (pass the device_code from Step 1)
lark-cli auth login --device-code "<device_code>" --json

Step 4: Write Actual Weight/Reps to Bitable

After extracting weight/reps data from comments, write it all to the Bitable (训练记录) with full detail for future analysis.

Bitable config:

• Base token: YOUR_BASE_TOKEN
• Table ID: YOUR_TABLE_ID ("训练记录")
• Primary field (记录标题): 笔记 (FLD_NOTE) — 必须作为 fields 第一个字段传入
• 记录标题命名规则:
• 运动打卡时: "{动作名} {计划}" (e.g., "杠铃颈前深蹲 4×6-8")
• 训练完成时: "{动作名} {实际重量汇总}" (e.g., "杠铃颈前深蹲 60-70×8 ✅")
• Fields (in order for batch create):
• 笔记 (text — 记录标题/主字段，必须放在第一个)
• 日期 (datetime): "2026-06-08 00:00:00" (YYYY-MM-DD format)
• 训练类型 (select): "Push A" / "Push B" / "Pull A" / "Pull B" / "篮球"
• 阶段 (text): e.g., "Phase 1: 神经唤醒与重组期"
• 动作 (text): Exercise name
• 计划 (text): e.g., "4×6-8"
• 重量 (text): Comma-separated per-set weights, e.g., "60,70,70,70"
• 次数 (text): Comma-separated per-set reps, e.g., "8,8,8,8"
• 完成 (checkbox): true/false
• 备注 (text): Summary of difficulty notes from comments

Batch create command:

#!/bin/bash
BASE_TOKEN="YOUR_BASE_TOKEN"
TABLE_ID="YOUR_TABLE_ID"
DATE="2026-06-08 00:00:00"
WORKOUT="Push A"
PHASE="Phase 1: 神经唤醒与重组期"

lark-cli base +record-batch-create \
  --base-token "$BASE_TOKEN" \
  --table-id "$TABLE_ID" \
  --json '{
    "fields":["笔记","日期","训练类型","阶段","动作","计划","重量","次数","完成","备注"],
    "rows":[
      ["杠铃颈前深蹲 60-70×8 ✅","2026-06-08 00:00:00","Push A","Phase 1: 神经唤醒与重组期","杠铃颈前深蹲","4×6-8","60,70,70,70","8,8,8,8",true,"第2组稍微吃力,第4组轻松"],
      ["保加利亚单腿蹲 15×8 ⚠️","2026-06-08 00:00:00","Push A","Phase 1: 神经唤醒与重组期","保加利亚单腿蹲","3×8-10/侧","15,15,15","8,8,8",true,"3组均吃力"]
    ]
  }' \
  --format pretty

Conventions for populating weight/reps/备注 from comments:

Sort comments by create_milli_time ASC to ensure correct set order.

Step 5: Analyze Completion

Compare actual completion against the plan:

Check each task's completed_at timestamp to verify it was marked done.

Step 6: Generate Analysis Report and Write Daily Report

Write the analysis as a Feishu Drive Markdown document using lark-cli markdown +create.

Folder structure: Create a "训练日报/YYYYMM/" folder hierarchy in Drive to organize reports.

6a. Create Monthly Folder (if not exists)

Use lark-cli drive +create-folder to create a monthly folder under the parent "训练日报" folder.

First, check if the "训练日报" folder and monthly subfolder exist:

# Search for existing "训练日报" folder to get its token
lark-cli drive +search --query "训练日报" --format pretty

If "训练日报" folder doesn't exist, create it:

lark-cli drive +create-folder --name "训练日报" --format pretty

Then create the monthly subfolder (e.g., "202606"):

lark-cli drive +create-folder --name "202606" --folder-token "<parent_folder_token>" --format pretty

Save the folder token from the response for use in the next step.

6b. Create Daily Report Document

Use lark-cli markdown +create with --content to create the daily report as a Markdown file in the monthly folder.

Since markdown content may contain Chinese characters and multiple lines, write the content to a local shell script file in the workspace, then execute it.

#!/bin/bash
lark-cli markdown +create \
  --name "训练分析 - Push A (2026-06-08).md" \
  --folder-token "<monthly_folder_token>" \
  --content '## 训练分析 - Push A (2026-06-08)

**完成情况**: ✅ 全部完成

**动作详情**:
| 动作 | 计划 | 完成 | 备注 |
|------|:----:|:----:|:------|
| 杠铃颈前深蹲 | 4×6-8 | ✅ | Phase 1 以轻重量保证动作质量 |
| 保加利亚单腿蹲 | 3×8-10/侧 | ✅ | 注意单腿稳定性 |
| 杠铃平板卧推 | 4×5-8 | ✅ | 重新建立神经肌肉连接 |
| 上斜哑铃卧推 | 3×8-10 | ✅ | 填补上胸厚度 |
| 坐姿哑铃肩推 | 3×8-10 | ✅ | 肩部立体感训练 |
| 绳索下压(三头) | 3×12-15 | ✅ | 辅助肌群 |
| 站姿提踵 | 3×15-20 | ✅ | 增强起跳启动 |

**营养执行**:
- 今日：高碳日
- 目标：2700kcal / 碳水300g / 蛋白150-160g / 脂肪70g

**当前阶段**: 神经唤醒与重组期（Week 1/4）
- 核心目标：轻重量专注动作质量，唤醒肌肉记忆

**改进建议**:
1. 后续在任务评论里记录实际重量和次数，方便追踪渐进超负荷
2. 明天如果有训练，注意补充碳水和热身

**鼓励语**: 干得漂亮！继续加油 💪
' \
  --format pretty

Note: Use --content for inline markdown content. For very long content, write the markdown to a local .md file first and use --file instead.

6c. Send Summary Message

After creating the document, send a summary message to the user via the message tool with the key findings and encouragement.

Include the document title and a brief note that the full analysis is now in their Feishu Drive.

Workflow 3: Quick Info Queries

When user asks "今天的训练是什么" / "今天吃什么" / "本周计划":

Today's Training + Nutrition

python3 scripts/get_fitness.py

Present a summary with:

• Workout type and focus
• Exercise list with sets/reps
• Nutrition (carb type, calories, macros)

Weekly Schedule

python3 scripts/get_fitness.py --week

Show the full week's plan.

Key Notes

• Script path: get_fitness.py is at scripts/get_fitness.py relative to skill root. The fitness.json data file is at the skill root.
• Calendar events: Create for ALL remaining training days when checking in, not just today.
• Task list vs standalone tasks: Always use +tasklist-create to bundle exercises into a task list, NOT +create for individual standalone tasks.
• Nutrition reminder: Always include phase context ("神经唤醒期 — 轻重量专注动作质量") and carb cycle info.
• Language: All task names, summaries, and communication should be in Chinese.
• Daily report: Use lark-cli markdown +create to write daily analysis to Feishu Drive, NOT local files. Create a "训练日报/YYYYMM/" folder structure in Drive to organize reports.
• Task comment API: Use Task v1 API (/open-apis/task/v1/tasks/{guid}/comments) to read comments. Task v2 does NOT support comment reading.
• Device flow re-auth: If a required scope (e.g., task:comment:read) is missing from the token:

1. Request: lark-cli auth login --device-flow --scope "task:task task:task:readonly task:comment:read" --format json
2. Ask user to visit the verification URL and enter the code
3. Complete: lark-cli auth login --device-code "" --json
• Bitable fields (训练记录): Full field list with IDs (字段顺序=API返回顺序, 笔记是主字段):
• FLD_NOTE: 笔记 (text — 记录标题/主字段，必须放在 fields 第一个)
• FLD_COMPLETED: 完成 (checkbox)
• FLD_PLAN: 计划 (text)
• FLD_REPS: 次数 (text — store as comma-separated values per set)
• FLD_REMARK: 备注 (text)
• FLD_DATE: 日期 (datetime, format: yyyy/MM/dd)
• FLD_WORKOUT_TYPE: 训练类型 (select: "Push A" / "Push B" / "Pull A" / "Pull B" / "篮球")
• FLD_WEIGHT: 重量 (text — store as comma-separated values per set)
• FLD_EXERCISE: 动作 (text)
• FLD_PHASE: 阶段 (text)
• Batch create field order (运动打卡): ["笔记","日期","训练类型","阶段","动作","计划","完成"]
• Batch create field order (训练完成): ["笔记","日期","训练类型","阶段","动作","计划","重量","次数","完成","备注"]
• 记录标题命名规则: 打卡时 "{动作名} {计划}", 完成时 "{动作名} {重量汇总}✅"