Where This Idea Came From

I asked Claude for a breakdown of what I actually use it for, by category because I wanted to know where I could expose it to more context to make it more useful. Claude looked back over our history and surfaced four recurring areas: Fulcra, another work project, my own creative work, and my house renovation. It pointed out that each new conversation about one of these started from scratch, and suggested a lightweight status file per project.

‍
Problem‍

Every time I opened a new conversation about "a white paper" or "a tracker," Claude either had to reconstruct context by searching, or ask me to re-explain which project, which person, which stage things were at. That wasn't just slow — it was a  source of mistakes. Claude had, at different points, misattributed the a white paper to the wrong author and mixed up which person was involved in which piece of work. Fixing that after the fact cost more time than it should have.

‍
Steps to Solve

• Asked Claude directly what I use it for most, by category, so we could see where the gaps actually were rather than guessing.
• Claude proposed one file per project — current stage, typical ask — tied together by a single index file it would check first. Rather than writing five files from scratch, had Claude pull forward what was already on file for two projects (Fulcra, Maine house) and write two new files for the gaps (other work, my own creative work) — so existing context didn't get duplicated or rewritten.
• Had Claude build the index last, linking out to each project file instead of repeating their content inside it.
• First pass, kept the whole tracker in Claude's own memory — fast to set up, but visible only to Claude.
• Asked whether it was worth writing into Fulcra too, since that's the layer other agents can actually see.
• Once the answer was yes, had Claude write the same index and four project files into Fulcra under /agents/, with Fulcra treated as the source of truth and Claude's memory copy kept as a synced pointer rather than a second, independent tracker.

‍

Solution

• Four project files, one per area, each holding only what's actually been established: what stage the project is in, what kind of ask typically comes up and a fifth index file that's the first thing any agent reads when picking up one of these threads — it points to the right project file instead of guessing which one applies.
• Started this in Claude's own memory, but moved it into Fulcra once it was clear that's where it actually belongs. Fulcra's whole premise is that any connected agent can pick up where another left off and a tracker of what's active and what stage it's at is the kind of context that shouldn't be stranded inside one AI product's memory.
• The project content itself (Fulcra's positioning doc, meeting recaps, renovation notes) was already in Fulcra; now the map of what exists and where to look lives there too, instead of split across two places.

‍

Result‍

Any connected agent — not just Claude — can now open a fresh conversation about one of these four areas and land on the right context without me re-explaining who's who or what stage we're at. The kind of misattribution that used to require a correction mid-conversation is a lot less likely, because the agent is reading from a file that's already been checked, not reconstructing from memory or guessing.

It's reusable: when a fifth project shows up, it gets its own file and one new line in the index — nothing else needs to change.

It draws a clean line: Fulcra is the source of truth for both the work itself and the map of what's active; Claude keeps a synced copy for quick access, but updates flow into Fulcra so nothing drifts out of sync between the two.