Launching today

Groundwork
Claude Code plans that re-runs never overwrite
1 follower
Claude Code plans that re-runs never overwrite
1 follower
A Claude Code skill that scaffolds a plan folder: research, design, plan, tracking, and an HTML board. Re-runs only rewrite the blocks it generated, checked by a hash. Your own text stays as you wrote it. Free, Apache-2.0.






Free
Launch Team

TinesThe single, secure environment for agents, apps, and automations.
Promoted
Hi, I'm Chinedum. I built Groundwork. I also run Royalti.io, a royalty platform for music companies. Groundwork came out of how I plan our own work.
The problem: I would write a plan with an AI agent, ask it to update the plan, and it would rewrite the whole file. That included paragraphs I had written by hand. The plan drifted a bit more each day, and I could not regenerate it without losing my notes.
Groundwork is a skill for Claude Code. You give it a goal and it makes a folder of numbered documents: plan, two research files, a discussion log, a tracking file, and an HTML board. Then you run small actions on it whenever you like: research, design, review, orchestrate.
What I think is different is one rule. Everything Groundwork generates sits inside a marked region. Everything outside is yours. A small Python script compares a hash of each region with what it last wrote. If you edited inside a region, it skips it and says so. The model does not decide that.
What it does today:
- Five profiles: software, general (campaigns, org changes), content, design-system, and film (pre-production).
- A board with Kanban, DAG and mission views. An explorer with search that works offline. A plans index across many plans.
- Stable IDs for work packages, gaps and designs, so a review knows what to update.
- orchestrate turns the plan into a wave order and a brief per work package. With --emit-workflow it writes a Claude Code Workflow script that runs the waves.
What it does not do:
- It does not write your plan. It keeps the folder in good shape.
- It needs Python 3 on your machine. The scripts use only the standard library.
- In a plain browser, the board's Start button copies a prompt to your clipboard. It does not start a session. Inside the Ikenga desktop app it can.
- The board loads React from a CDN the first time you open it, so it needs a network. The explorer does not.
- The research action writes web content into a file. Treat that as untrusted text and read it before you hand it to agents.
It is free and open source under Apache-2.0. You do not need Ikenga, our desktop app, to use it.
I would like to hear where it breaks, especially on work that is not code. I will answer every comment today.