skip to content
carldmitchell.com
← all projects

IdeaBoard

Where my ideas start. Every idea is a folder with its own README and CLAUDE.md, so I can open a Claude session inside any idea and run research or a critique on it, then promote it into a real project.

details

status
in progress
group
ai tooling
stack
Python (standard library only), plain JS, Markdown with frontmatter; the window is a second instance of Switchboard
since
2026-06

readme

spark โ†’ seed โ†’ defined โ†’ promoted to a project    (+ parked, graveyard)

Switchboard, my project switcher, is where my real projects live. IdeaBoard is the league below it: an idea works its way up here and, once it's defined and worth building, its folder moves into one of Switchboard's roots and becomes the project.

An idea is a folder

Each idea is a folder holding a README.md (the card: what it is, why now, the open questions) and a CLAUDE.md (the rules for any session started inside it). Research passes and critiques land beside them as dated files. The stage: line in the README's frontmatter is the only thing the window reads to place an idea, and moving an idea rewrites only that word. The files are the source of truth, so the board is greppable and editable by hand or by an agent with no API in between.

The window

The IdeaBoard window is Switchboard itself, run a second time with its own config: one codebase, two instances. It lists ideas from three roots, grouped Defined, Seed and Spark from top to bottom, so an idea climbs the list as it nears Switchboard. From there I can:

  • Start a session inside any idea.
  • Run Research or Critique. Each button starts a session in the idea if it has none and sends a prepared prompt. Research looks for who already does it, how real the need is and what it would take. Critique makes the strongest case against it and ends with a verdict: drop, park, sharpen or go.
  • Promote a defined idea: pick a Switchboard root and a name, and the folder moves there with its research and its session history. A one-file pointer stays behind, so the record of where it went is on disk.

How ideas get in

  1. Automatically: my funnel skills (see Claude skills) scout trend signals, score them, and land the best as spark folders in the right root.
  2. By hand: a "new idea" button that fills both files from a template, or I ask Claude to add one.

Details I like

  • One home per subject. General ideas live here; music and Suno ideas live inside their own repos, and the window reads all three roots.
  • Sessions research, I decide. An idea's CLAUDE.md tells a session to write research and critiques, not to build the thing or change its stage. Moving an idea is my call.
  • Leads aren't ideas. News leads and research briefs stay on the original file-backed kanban, a small standard-library Python server with a Mailroom and a Research space.

59 idea folders across the three roots, 14 of them pointers to ideas already promoted into projects.