Inspiration To-do lists are boring, and Halloween is the one season where boring is not allowed. We wanted a task list that feels alive — or undead. In Tombstone Tasks every task is a tombstone in a graveyard: you plant it, it stands among the dead, and when you finish it you bury it with an epitaph. Overdue tasks don't just turn red — they rise from the dead.
What it does Add a task with title, priority, and due date; it appears as a tombstone in the graveyard. Bury a finished task with an epitaph (write your own, or skip and one is auto-generated). Overdue tasks become "risen": fog, cracks, and a warning, so you can't ignore them. Filters (All / Alive / Buried), stats (Alive, Buried, Streak in survival days), and a moon in the header that brightens as your streak grows. Works fully offline: no build step, no dependencies, no server, zero network requests. Tasks persist in localStorage. A Playwright smoke test (load, add, bury, filters, risen) passes 5/5.
How we built it Spec-driven development with Kiro (CLI 3.5.0). The evidence is in the repo's .kiro/ folder: .kiro/specs/tombstone-tasks/ — requirements.md (EARS requirements), design.md, tasks.md (27 implementation tasks in 4 dependency waves); .kiro/steering/ — persistent project knowledge loaded into every session; .kiro/hooks/ — a smoke-test-on-save hook that runs the Playwright smoke test after every save; .kiro/agents/ — a graveyard-builder custom agent used for the build. The app code was authored inside Kiro sessions executing the spec tasks wave by wave (26 sessions, one file per run in small chunks, runs spaced out), with the smoke test verifying each wave. The demo video and this submission were prepared as logistics around that Kiro-built code — no app code was written outside Kiro.
Challenges we ran into Kiro's API throttled us mid-build (HTTP 429, credit-consumption rate limit) right in the middle of Wave 1. We waited out the hour, then resumed with the same discipline that got us through the rest: one file per run, small chunks, at least 10 minutes between runs, and verifying on disk after every session before retrying. A spec contradiction also surfaced late (seed data vs empty-init requirement) — we implemented the spec as written and flagged it rather than silently changing it.
What we learned Spec-first beats prompt-first: with requirements, design, and tasks written before any code, every Kiro session knew exactly what to build and the smoke test could verify it. Small, spaced-out sessions survive rate limits that kill big hammering runs.
What's next Sound effects (a bell toll on burial), shareable graveyard snapshots, and a "haunted streak" calendar.License: https://github.com/agusher21/tombstone-tasks/blob/main/LICENSE
Built With
- css3
- html5
- javascript
- kiro
- playwright
