Smart Tasks

Intelligent task management using workspace markdown files. Natural language CRUD via chat, context-aware reminders, priority briefings, and weekly reviews. Use when: user mentions tasks, deadlines, todos, or things to do; user asks "what should I do today/this week"; user wants to track, complete, or review work items. NOT for: calendar events, meeting scheduling, or project management with multiple collaborators.

Manage tasks as markdown files in tasks/ with an INDEX.md summary. Minimize context usage: read INDEX.md first, individual files only when needed.

Setup

Run the init script once to create the directory structure:

bash skills/smart-tasks/scripts/init-tasks.sh

This creates tasks/, tasks/active/, tasks/done/, tasks/archive/, tasks/INDEX.md, and tasks/.meta.json. Idempotent — safe to re-run.

After init, apply workspace integration changes from workspace-integration.md with user approval.

File Structure Overview

• tasks/INDEX.md — active task summary table (~1-2KB). Read every session.
• tasks/.meta.json — tracks nextId counter and known categories.
• tasks/active/ — individual task files (one per task).
• tasks/done/ — completed tasks (kept 30 days).
• tasks/archive/YYYY-QN/ — long-term storage.

For full format specs see file-structure.md.

Core Workflow

1. Detect Task Intent

When the user mentions deadlines, todos, or things to do in conversation, propose creating a task. No special commands needed — use natural language.

Examples of triggers:

• "다음 주 금요일까지 리뷰 완료해야 해"
• "논문 피드백 보내야 하는데"
• "이번 주 할일 뭐 있어?"

2. Create a Task

1. Read tasks/.meta.json to get nextId.
2. Generate ID: T-{nextId} zero-padded to 3 digits (e.g., T-001).
3. Create slug from title (kebab-case English summary).
4. Write task file to tasks/active/{ID}_{slug}.md using the format in task-format.md.
5. Fill the Agent Notes section with relevant workspace context (related files, people, timeline analysis).
6. Increment nextId in .meta.json. Add new category if not already listed.
7. Update tasks/INDEX.md — insert row in the correct section.

3. Update a Task

1. Identify the task by ID or title search in INDEX.md.
2. Read the individual task file from tasks/active/.
3. Apply changes to both the task file and INDEX.md.
4. Append a timestamped entry to the task's 메모/진행 기록 section.

4. Complete a Task

1. Set status: done and completed: {today} in the task file.
2. Move file from tasks/active/ to tasks/done/.
3. Remove the row from INDEX.md. (Do NOT add to done section in INDEX.)
4. Confirm completion to the user.

5. Query Tasks

• Quick overview: Read INDEX.md only. Answer from the table.
• Detail needed: Read the specific task file(s) from tasks/active/.
• "What should I do?": Analyze INDEX.md by deadline × priority × estimated hours. Recommend a sequence with reasoning.

6. Delete a Task

Only when explicitly requested. Move to tasks/done/ with status: cancelled rather than deleting the file. Remove from INDEX.md.

INDEX.md Management

INDEX.md is the single-read entry point for task awareness.

Structure

# Tasks Index

> Last updated: YYYY-MM-DD
> Active: N | Overdue: N | Due this week: N

## 🔴 Overdue
| ID | Title | Due | Priority | Category |
|----|-------|-----|----------|----------|

## 🟡 Due This Week
| ID | Title | Due | Priority | Category |
|----|-------|-----|----------|----------|

## 📋 Later
| ID | Title | Due | Priority | Category |
|----|-------|-----|----------|----------|

## 📊 Summary

Update Rules

• Update INDEX.md on every task create / update / complete / delete.
• Recalculate header counters (Active, Overdue, Due this week) each update.
• Keep rows sorted by due date within each section.
• Reclassify rows between sections based on current date.
• Set Last updated to today.

Size Control

Active tasks	~Size	Action
1–20	~1KB	Keep as-is
20–50	~2.5KB	Abbreviate Later section (title + due only)
50+	~5KB+	Split into category indexes (INDEX-{category}.md)

Context Strategy

Read context-strategy.md for details on:

• Layered reading pattern (INDEX → active file → done/archive)
• Subagent cost optimization for cron jobs
• When to read vs. skip individual files

Key rule: If INDEX.md alone can answer the question, do NOT read individual files.

Cron Jobs

Set up scheduled briefings and alerts. All cron jobs use Sonnet-class models. See cron-templates.md for:

• Daily morning briefing (08:00) — overdue alerts + today's priorities
• Deadline check (12:00, 18:00) — D-0/D-1/overdue only, silent if none
• Weekly review (Sunday 20:00) — completion summary, stale task detection, archive cleanup (done/ 30+ days → archive/)

Automatic Cleanup

Transition	Trigger	Action
active → done	User marks complete	Move file, update INDEX
done → archive	30 days elapsed	Weekly review moves to archive/YYYY-QN/
archive	—	Never in INDEX. Access only via search.

Task File Format

Each task is a markdown file with YAML frontmatter. See task-format.md for:

• Complete frontmatter schema
• Section template
• 2-3 example tasks

Workspace Integration

This skill does NOT modify files outside its own directory. See workspace-integration.md for recommended changes to AGENTS.md, SOUL.md, and HEARTBEAT.md (diff format). Apply these with user approval during setup.

Integration with Other Skills

• proactivity: Record task status in session-state; proactively surface relevant tasks during conversation.
• self-improving: Log task management mistakes/improvements to domain file. Create self-improving/tasks.md for task-specific learnings.

Best Practices

1. Always update INDEX.md and the task file atomically.
2. Fill Agent Notes with cross-references to workspace files at creation time.
3. Use estimated_hours to give realistic scheduling advice.
4. When reminding about tasks, include relevant context — not just the deadline.
5. Categories are user-defined. Never hardcode a fixed set.
6. Keep task titles concise but descriptive for INDEX readability.