3h • 🛠️ Show Your Work
Something I built this week that will help anyone working with Claude Code
Background first — because how this came together matters more than the file itself.
---
## The problem I kept running into
Every time I started a Claude Code session, I was re-explaining everything.
Who I am.
What I’m building.
What matters.
What not to touch.
Session after session.
The context would bloat.
The model would drift.
And I’d end up managing the AI instead of working with it.
I thought the fix was a better prompt.
It wasn’t.
---
Ari introduced me to the dispatch pattern.
Instead of relying on chat history, you write a brief to disk before the session starts.
The brief is the contract.
The worker (Claude) opens it cold, executes, and returns a result.
No shared history.
No “as we discussed.”
No hidden context.
Her work on:
- structured briefs
- worker isolation
- and the cold-start rule
completely changed how I think about AI.
The cold-start rule was the shift:
> If the brief cannot stand alone, it is not finished.
---
## What @JakeVanClief framework gave me
Here I learned to think in systems instead of prompts.
Three files.
That’s enough for any new session to orient itself quickly, without scanning everything.
That’s the foundation.
---
## The realization
These aren’t different ideas.
They’re the same idea at different scales.
- A worker sandbox is a dispatch worker
- A context file is a routing layer
- A system architecture is just structured dispatch
Once that clicked, the gap became obvious.
---
## So I built the missing layer
A bootstrap file.
One markdown file that sits at the top of a project and does what neither system does on its own:
It aligns:
- the worker model
- the architecture
- and the rules
**before any work begins**
---
## What it actually does
Drop it into a project, and Claude:
- understands the system
- checks what already exists
- configures itself
- and only then starts working
No setup.
No re-explaining.
No drift.
---
## The part most people miss
The most expensive AI mistakes aren’t bad outputs.
They’re silent assumptions.
So the bootstrap includes hard guardrails — not suggestions, actual stop conditions — that apply before any task runs.
That’s what prevents rework.
---
## The file
It’s built for my world:
- yoga brand
- books
- Kajabi funnels
- content systems
But the pattern is universal.
You change:
- brand section
- project specifics
Everything else stays.
The underscore prefix is intentional — it forces the file to the top of any folder.
First thing Claude sees.
---
## Credit where it’s due
Ari — the dispatch pattern, the brief structure, the cold-start rule.
That’s hers.
Jake — the architectural system that shifted me from prompts to systems.
That came through his work and the community built around it.
---
## Final thought
If you’re still re-explaining your project every session, this isn’t a prompt problem.
It’s a system problem.
This is the system.
— Gabe
2
4 comments
Something I built this week that will help anyone working with Claude Code
skool.com/quantum-quill-lyceum-1116
Jake Van Clief, giving you the Cliff notes on the new AI age.
Leaderboard (30-day)
1
+732
2
+562
3
+465
4
+406
5
+386
Powered by