GTD Project

Reference: See things3 skill for task requirements, tag taxonomy, and MCP operations.

Work through a Things 3 project by claiming and completing its child tasks sequentially. Use the project as a work bucket for discovered items, review feedback, and follow-ups.

Arguments

The skill receives a project name or search term as $ARGUMENTS. Example:

• /gtd:project Retool around Things 3
• /gtd:project user authentication

Workflow

1. Find the Project

Search for the project:

mcp__things__search_items(query="$ARGUMENTS")

If multiple matches, list them and ask which one. If one match, proceed.

2. Get Project Tasks

Retrieve all tasks in the project:

mcp__things__get_todos(project_uuid="<project-uuid>", include_items=true)

Report the project overview:

Project: [name]
Tasks: [count] total, [incomplete] remaining

Tasks:
1. [ ] First task [tag]
2. [ ] Second task [tag]
3. [x] Completed task

3. Work Through Tasks In Order

IMPORTANT: Tasks are returned in the exact order the user arranged them in Things 3. Always work through tasks in this order - start with the first incomplete task, complete it, then move to the next. Never skip ahead or reorder based on your own judgment.

For each incomplete task (in order):

1. Claim the task:

   mcp__things__update_todo(id, tags=["Working"])
   

2. Analyze the task:

   • Read task notes for context
   • Check if it's Agent/Queued or human-required
   • Determine if Claude can execute

3. Execute or assist:

   • AI-automatable: Research, draft, implement - just do it
   • Human-required: Provide context, then wait for confirmation

4. Report progress:

   Working on "Task title":
   - [x] Step 1 (done)
   - [ ] Step 2 (in progress)
   

5. Complete the task:

   mcp__things__update_todo(id, completed=true, tags=[], notes="[existing + completion summary]")
   

6. Re-fetch tasks and continue:

   mcp__things__get_todos(project_uuid="<project-uuid>", include_items=true)
   

   Always re-fetch the task list after completing a task. The user may have reordered or added tasks while you were working. Start again from the first incomplete task in the fresh list.

4. Handle Discovered Work

When you discover new work while working on a task:

mcp__things__add_todo(
    title="New task discovered",
    notes="Context: Found while working on [original task]",
    tags=["25m"],
    list_id="<project-uuid>"
)

5. Update Project Notes

Track progress and decisions in the project description:

mcp__things__update_project(
    id="<project-uuid>",
    notes="[existing notes]

---
## Agent Progress
- Completed: [tasks done]
- Discovered: [new items added]
- Blocked: [any blockers]
- Notes: [key findings or decisions]"
)

6. Project Completion Rules

NEVER mark the project complete. Only the human decides when a project is truly done.

When all visible tasks are complete, tag the project for review:

mcp__things__update_project(
    id="<project-uuid>",
    tags=["Agent", "Needs Review"],
    notes="[existing notes]

---
## Ready for Review
- Completed: X tasks
- Added: Y new tasks discovered during work
- Summary: [brief description of what was accomplished]

Please review and either close the project or add follow-up tasks."
)

Then report to the user:

All current tasks in "[Project Name]" are complete.

Summary:
- Completed: X tasks
- Added: Y new tasks

Tagged project as "Agent/Needs Review" for your review.
Please add follow-up tasks or close the project when satisfied.

MCP Tools Reference

Guidelines

• Respect task order - Work tasks in the exact order returned by the API (this matches the user's arrangement in Things 3). Never skip or reorder tasks.
• Re-fetch after each task - Always get a fresh task list after completing a task. The user may reorder or add tasks while you work.
• Claim before working - Always update tags to "Working" first
• Use project as bucket - Add discovered work to the project, not elsewhere
• Update project notes - Document progress, decisions, and findings
• NEVER close projects - Only humans decide when projects are done
• Be autonomous - Execute AI-automatable tasks without asking
• Report progress - Keep the user informed of what you're doing
• Ask when stuck - If blocked, explain why and ask for guidance