The Problem
I use Todoist to manage everything. When a task feels too big or unclear, I tag it with an overwhelmed label so I can come back to it. The problem is that “come back to it” usually means staring at the same vague task a week later, still not knowing where to start.
I already had a /overwhelmed command in Claude Code that acted as a scope-cutting coach. It would ask me questions to narrow down the next best move. It worked well for general overwhelm, but it was disconnected from my actual task list. I would have a conversation with Claude, figure out the pieces, and then manually go create subtasks in Todoist. That last step added enough friction that I often skipped it.
The Idea
I wanted the skill to do three things:
- Pull tasks tagged
overwhelmeddirectly from Todoist - Use Socratic questioning to help me figure out the breakdown myself
- Write the subtasks back to Todoist when we’re done
The Socratic part matters. I tried approaches where Claude proposes the breakdown for you, but the whole reason a task feels overwhelming is that you haven’t thought it through yet. Having an AI suggest subtasks skips the part that actually helps. You end up rubber-stamping a breakdown that doesn’t match your real situation.
How It Works
When I run /overwhelmed or tell Claude I’m stuck, it fetches everything tagged overwhelmed from Todoist using the MCP connector. Right now that’s things like “Learn: k8s theory” and “Performance testing of Kafka cluster.” It shows me the list and asks which one I want to tackle.
Then the questioning starts. One question at a time, no batching:
- “What specifically about this feels like too much?”
- “What would the ‘next best thing’ look like for this? Not perfect, not complete — just the next best thing.”
- “What’s the first thing you’d need to figure out or do?”
- “Could you sit down and start this without needing to figure anything else out first?”
If I start expanding scope, it pushes back. If I’m stuck, it asks what I’m avoiding. These questions are not original — they are drawn from coaching frameworks. But having them asked at the right moment, in the context of a specific task I’m procrastinating on, is what makes them useful.
Once we’ve drawn out all the pieces, Claude summarizes them using my own words and asks me to confirm. Then it creates subtasks under the original task in Todoist and removes the overwhelmed label. The parent stays completable. The subtasks start with a clean slate — no inherited priorities or due dates unless I explicitly set them during the conversation.
If there are no overwhelmed tasks in Todoist, it falls back to the original scope-cutting coach behavior. Ask what’s feeling like too much, narrow it down, end with “Your next best thing is:” and offer to add it to Todoist.
The Design Process
I built this by brainstorming with Claude before writing anything. We went through every decision one question at a time: Should subtasks inherit parent properties? (No, start fresh.) Should the AI propose breakdowns? (No, pure Socratic.) Should it replace the old command or live alongside it? (Replace it, fold the best parts in.)
The result is a single markdown file at ~/.claude/commands/overwhelmed.md. No code, no dependencies. Just a prompt that instructs Claude on the conversational flow and which Todoist MCP tools to call at each step.
Takeaway
The gap between knowing what you need to do and being able to start is usually a decomposition problem, not a motivation problem. Tasks feel overwhelming because they’re vague, not because they’re hard. Having a tool that forces you to articulate the pieces — without doing the thinking for you — is more useful than any AI-generated task breakdown.