Plan Interview: turn vague goals into actionable Todoist projects

Plan Interview is an Agent Skill for people who know what they want to accomplish but cannot see the next concrete step. It runs a short, obstacle-aware interview, drafts a small milestone plan for approval, and only then writes the approved project to Todoist.

Install

Canonical skill URL:

https://github.com/Loffee5422/plan-interview-skill/tree/main/skills/plan-interview

Paste this instruction into a compatible agent:

Install the `plan-interview` skill from https://github.com/Loffee5422/plan-interview-skill/tree/main/skills/plan-interview

With Codex's bundled GitHub skill installer:

python3 "${CODEX_HOME:-$HOME/.codex}/skills/.system/skill-installer/scripts/install-skill-from-github.py" \
  --url https://github.com/Loffee5422/plan-interview-skill/tree/main/skills/plan-interview

For reproducible installation, pin the initial release:

https://github.com/Loffee5422/plan-interview-skill/tree/v1.0.0/skills/plan-interview

What it does

• Converts a broad outcome into 5–12 concrete actions for the next milestone.
• Keeps each task small enough for a focused 15–45 minute work session.
• Identifies the real obstacle and attaches an if–then response to it.
• Creates an intentionally tiny ignition task and exactly one visible next task.
• Distinguishes genuine deadlines from personal scheduling.
• Shows the complete Todoist draft and requires explicit approval before any write.
• Revisits stalled tasks as plan-design problems and breaks them down further.

Representative uses

• Break a report, application, launch, or study goal into next actions.
• Plan the next week of a larger project without generating an overwhelming backlog.
• Create a Todoist project with sections, labels, estimates, and hard deadlines.
• Review an existing project and repair tasks that have remained untouched.

Requirements and compatibility

The skill is written for agents that support the Agent Skills SKILL.md format. Todoist writes require an available, authenticated Todoist MCP integration whose live tool schema the agent can inspect. Without Todoist access, the interview and approved plan still work in chat as a copy-pasteable fallback.

Task-duration support varies by Todoist plan and integration. When a duration field is unavailable, the skill directs the agent to put the estimate in the task description.

How it works

The workflow has four phases:

1. Interview briefly for the outcome, real deadline, current state, available time, and likely obstacle.
2. Draft a compact plan and revise anything that still feels vague or too large.
3. After explicit approval, create the Todoist project, sections, labels, and tasks in dependency order.
4. Close the loop with the one task to start and the most relevant if–then plan.

The operational instructions live in skills/plan-interview/SKILL.md.

Example

User:

I keep putting off my research proposal. Help me make a Todoist plan.

The skill asks only the next one or two useful questions at a time, drafts a small milestone plan, and asks whether any task still feels vague. It does not touch Todoist until the user explicitly approves the displayed structure and exact write scope.

FAQ

Does this skill create tasks immediately?

No. It must display the complete draft and receive a clear approval before creating or changing anything in Todoist.

Does it schedule every task?

No. It uses deadline fields only for genuine constraints. The user remains in control of day-to-day scheduling.

Can I use it without Todoist?

Yes. The agent can complete the interview and return a clean plan in chat. Only the write phase requires Todoist.

Is this a full project-management system?

No. It deliberately plans only the next milestone or roughly the next week. It is designed to make starting easier, not to produce a comprehensive long-range backlog.

What happens when a task stalls?

Review mode checks what actually happened, treats persistent non-starting as evidence that the task may be too vague or large, and creates a smaller ignition step.

Limitations

• The quality of the plan depends on the user's answers about constraints and obstacles.
• The skill does not infer deadlines, make calendar commitments, or bypass Todoist approval.
• Todoist capabilities and field names are discovered at runtime and may differ by integration or account.
• It is a planning workflow, not medical or mental-health treatment.

Contributing

See CONTRIBUTING.md for the contribution workflow. Please keep changes focused on making tasks concrete, small, safe to write, and easy to begin.

License

Released under the MIT License.