Daily Workflow (开工啦)

SKILL.md

---
name: daily-workflow
description: "Maintain project work-session status before, during, and after work. Use when the user says Chinese or English trigger phrases such as 开工啦/start work, 中段检查/checkpoint, 吃饭啦/lunch check, 收工啦/wrap up, or 交接/handoff, and needs Docs/ project notes updated for user visibility and AI-to-AI handoff."
---

# Daily Workflow Skill / 日常工作流 Skill

Version: 2.0.0
Purpose: keep a project understandable across work sessions and across AI assistants.
Storage: project-local `Docs/` only.

## Core Principle / 核心原则

Write concise, factual project state that a user can scan and the next AI can continue from. Prefer concrete status, blockers, files, commands, and next actions over narrative.

Do not record secrets, credentials, private customer data, or large sensitive source excerpts.

## Context Compression and Reset / 上下文提炼与清理

At checkpoint, wrap-up, and handoff time, compress the current conversation context before updating workflow files.

Context compression means extracting only the information needed to continue work:

- Current state and user intent
- Decisions already made
- Completed work
- Pending work
- Blockers, risks, and assumptions
- Files touched or important paths
- Commands/tests run and their outcomes
- Immediate next action

Write this compressed context into `Docs/STATUS.md` under the latest update entry. Do not paste the full conversation transcript.

After saving the compressed context, treat the old working context as cleared for workflow purposes. Continue from the compressed context in `STATUS.md` and the action plan in `SCHEDULE.md`, not from long prior chat history.

For checkpoint only, provide the compressed context back to the current AI in the response so work can continue immediately from the reduced context.

## Trigger Phrases / 触发语

Treat these phrases as equivalent triggers:

- Start work / 开工: `开工啦`, `开始工作`, `start work`, `begin work`
- Mid-session checkpoint / 中段检查: `中段检查`, `吃饭啦`, `checkpoint`, `lunch check`, `midday check`
- Wrap up / 收工: `收工啦`, `结束工作`, `wrap up`, `end work`
- Handoff / 交接: `交接`, `handoff`, `next AI`, `下一任 AI`

If the user uses custom triggers documented in `Docs/CONFIG.md`, honor them.

## Configuration / 配置

Optional configuration is project-local:

```text
Docs/CONFIG.md
```

Default config block:

```json
{
  "version": "2.0.0",
  "startPhrase": "开工啦",
  "checkpointPhrase": "中段检查",
  "lunchPhrase": "吃饭啦",
  "endPhrase": "收工啦",
  "handoffPhrase": "交接",
  "language": "bilingual",
  "firstRun": false
}
```

Create or update this config only inside `Docs/`. Do not write global user state.

## Managed Files / 管理文件

Create `Docs/` if it is missing. Maintain these files:

- `Docs/TARGET.md`: project purpose, success criteria, and scope. Default read-only after creation unless the user changes the goal.
- `Docs/STATUS.md`: current state, progress, recent updates, and known risks. Update the current status section while preserving history.
- `Docs/COMPLETED.md`: completed work log. Append only; never overwrite older entries.
- `Docs/PENDING.md`: current pending work, blockers, and decisions needed. May be rewritten as the current queue, but preserve unresolved blockers.
- `Docs/SCHEDULE.md`: immediate continuation plan and handoff block. Overwrite each checkpoint with the latest actionable next steps.

Optional file:

- `Docs/HANDOFF.md`: create only when the user asks for a standalone handoff file or when `SCHEDULE.md` becomes too long.
- `Docs/CONFIG.md`: optional trigger and language preferences.

## File Templates / 文件模板

### TARGET.md

```markdown
# Project Target / 项目目标

## Purpose / 目的
[What the project is trying to accomplish.]

## Success Criteria / 成功标准
- [Criterion 1]
- [Criterion 2]

## Scope / 范围
In scope:
- [Item]

Out of scope:
- [Item]

## Last Reviewed / 最后确认
[YYYY-MM-DD]
```

### STATUS.md

```markdown
# Project Status / 项目状态

## Current State / 当前状态
[Short factual summary.]

## Progress / 进度
[Percent or milestone-based progress when known.]

## Active Context / 当前上下文
- Files or areas being worked on:
- Important constraints:
- Known risks:

## Compressed Context / 提炼上下文
- User intent:
- Decisions made:
- Current state:
- Completed work:
- Pending work:
- Blockers / risks:
- Files touched:
- Commands/tests run:
- Immediate next action:

## Update History / 更新历史
### [YYYY-MM-DD HH:mm] [start/checkpoint/wrap-up/handoff]
- [What changed]
- [Evidence, command, or file reference when useful]
```

### COMPLETED.md

```markdown
# Completed Jobs / 已完成工作

## [YYYY-MM-DD]
- [Completed item with enough detail for another AI to understand what changed.]
```

### PENDING.md

```markdown
# Pending Jobs / 待办工作

## Immediate / 立即处理
- [ ] [Task, owner/context if known, acceptance condition]

## Later / 稍后处理
- [ ] [Task]

## Blockers and Decisions / 阻塞与待决策
- [Blocker or decision needed]
```

### SCHEDULE.md

```markdown
# Next Steps / 下一步

## Immediate Next Action / 下一步立即行动
1. [Action with file/path/context]
2. [Action]

## Handoff Summary / 交接摘要
- Current state:
- Completed since last checkpoint:
- Pending work:
- Blockers / risks:
- Files touched:
- Commands/tests run:
- Immediate next action:

## Notes for Next AI / 给下一任 AI 的说明
[Any context the next AI must know before acting.]
```

## Start Work Flow / 开工流程

When triggered by start work:

1. Check `Docs/`; create missing managed files from templates.
2. Check `Docs/CONFIG.md`; honor it if present, but do not require it.
3. Read all managed files.
4. Summarize the current project state to the user: target, current status, recent completed work, pending work, and immediate next action.
5. If the target is missing or unclear, ask the user for the project goal before writing `TARGET.md`.
6. Update `STATUS.md` with a start entry.
7. Refresh `SCHEDULE.md` only if the existing next steps are missing, stale, or contradicted by the user.

## Checkpoint Flow / 中段检查流程

When triggered by checkpoint or lunch check:

1. Compress the current conversation context into a concise continuation summary.
2. Append completed items to `COMPLETED.md`.
3. Update `STATUS.md` with current state, progress, risks, commands/tests run, and the compressed context.
4. Rewrite `PENDING.md` as the current queue while preserving unresolved blockers.
5. Rewrite `SCHEDULE.md` with a complete handoff block.
6. Treat the previous long context as cleared for workflow purposes.
7. Return the compressed context to the current AI in the response, followed by the immediate next action.

## Wrap Up Flow / 收工流程

When triggered by wrap up:

1. Compress the current conversation context into a concise end-of-session summary.
2. Append all completed work to `COMPLETED.md`.
3. Update `STATUS.md` with final state, commands/tests run, risks, and the compressed context.
4. Update `PENDING.md` with unfinished work, blockers, and decisions needed.
5. Update `TARGET.md` only if the goal changed or the user explicitly asks.
6. Rewrite `SCHEDULE.md` as the main handoff document.
7. Treat the previous long context as cleared for workflow purposes.
8. Give the user a short wrap-up summary and the immediate next action for a future session.

## Handoff Flow / 交接流程

When triggered by handoff:

1. Read the managed files.
2. Compress the current conversation context into a concise handoff summary.
3. Update `STATUS.md` with the compressed context and latest observed state.
4. Produce or refresh the `SCHEDULE.md` handoff block.
5. Include only actionable context: current state, completed work since last checkpoint, pending tasks, blockers and risks, touched files, commands/tests run and outcomes, and immediate next action.
6. Treat the previous long context as cleared for workflow purposes.
7. Do not invent results that were not observed.

Create `Docs/HANDOFF.md` only if the user asks for a separate handoff artifact.

## Write Rules / 写入规则

- Use absolute paths when reporting file locations to the user.
- Append to `COMPLETED.md`; do not overwrite it.
- Preserve `STATUS.md` update history.
- Preserve unresolved blockers when rewriting `PENDING.md`.
- Overwrite `SCHEDULE.md` at each checkpoint or wrap-up so the next AI has a single current plan.
- Never delete managed files automatically.
- Do not edit files outside `Docs/` for this workflow unless the user's broader task requires it.

## Safety Rules / 安全规则

Do not write:

- API keys, tokens, passwords, cookies, private keys, or `.env` values.
- Full private customer records or confidential source text.
- Browser session details or account credentials.
- Large unrelated logs.

If important context contains sensitive details, summarize it safely and name the secure source location only when appropriate.

## User Response Style / 用户反馈格式

Keep responses short and useful:

```text
Saved workflow state.
- Updated: STATUS.md, SCHEDULE.md
- Appended: COMPLETED.md
- Pending: 3 items, 1 blocker
- Next: run the failing test and inspect the parser change
```