A personal achievement partner disguised as a todo app. It doesn't manage tasks — it helps humans do things by working with their psychology, not against it.
Imperfect action beats perfect planning. The whole point of showing three doors is to get someone moving — not to optimize their todo list. Every design decision should reduce friction to starting, not add friction for "correctness."
Choice paralysis is real. Overwhelm is real. The "I'll do it later" trap is real. ThreeDoors exists because traditional todo apps ignore these realities. Features should acknowledge that humans are not rational task-processing machines.
Show 3 tasks. Not 5. Not "all of them with filters." The constraint IS the feature. When in doubt, show less. Resist the urge to add "just one more option."
- Data stays on the user's machine. Period.
- No telemetry, no analytics, no phone-home.
- No accounts, no sign-up, no cloud sync (unless the user explicitly configures it).
- Integrations (Apple Notes, Jira, GitHub, etc.) use local APIs or user-provided tokens — never intermediary services.
People already have tasks in Apple Notes, Jira, Linear, text files. ThreeDoors integrates with existing tools — it doesn't ask users to migrate. The adapter pattern exists precisely for this: plug into what people already use.
This is one human directing a team of AI agents. The human has limited hours per week; the agents don't. But complexity still costs — every feature must justify itself because the human must review, decide, and course-correct. Prefer the simple solution that works today over the elegant solution that takes three sprints. If a feature requires more than one story to be useful, reconsider the decomposition.
This project is also an experiment in human-agent team collaboration. The agent governance infrastructure (watchdogs, envoy, party mode, decision tiers) is as much a part of the project as the TUI itself. Lessons learned here inform how future projects are structured.
The human maintainer's primary job is decision-making, quality review, and direction-setting — not implementation. With agents handling 96% of PRs, the bottleneck is never "can we build it fast enough?" — it's "are we building the right thing?" Agent infrastructure should minimize the decisions that require human attention, not maximize the work agents can do.
This project was dormant for 4 months (Nov 2025 — Mar 2026). What unlocked velocity wasn't better tools — it was adopting BMAD as a methodology. The framework provides the structure that lets agents work autonomously while keeping the human in control of direction.
When implementing a story and you face a decision not covered by the spec:
- Would this reduce friction? If yes, do it. If it adds a step for the user, don't.
- Is this the simplest thing that works? Ship that. Refactor later if needed.
- Does this respect the user's data? No silent writes, no data loss, atomic operations.
- Does this follow existing patterns? Check how similar things are done in the codebase. Don't invent new patterns when existing ones work.
- Would the user notice this? If it's an internal refactor with no visible change, keep the scope minimal. If it's user-facing, make it feel effortless.
- Not a project management tool (no Gantt charts, no sprints, no team features)
- Not a habit tracker (no streaks, no gamification, no guilt)
- Not a second brain (no knowledge graph, no linking, no tagging taxonomy)
- Not trying to be everything to everyone — it's a personal tool for one person at a time
Keypresses should produce visible, satisfying responses. The UI should feel like physical objects — doors that open, selections that click into place. Subtle is not the same as invisible. When the user acts, the UI should answer unmistakably.
The difference between "I clicked a flat screen" and "I pressed a physical button" is the difference between adequate and delightful. We want the button feel.
Opening ThreeDoors should feel like a friend saying: "Hey, here are three things you could do right now. Pick one. Any one. Let's go."
Not: "You have 47 overdue tasks. Here's a productivity report."