Free build guide
A 6-file documentation system that gives Claude Code persistent memory across sessions. Stop re-explaining your stack. Stop re-discovering decisions you already made. Start every session like the last one never ended.
PDF · No email required · Built for Claude Code & Cowork
The problem
You open a fresh Claude Code session. You spend the first twenty minutes re-loading context: "here's our stack, here's why we chose Supabase over Firebase, here's the migration we just finished, here's the three things the last session tried that didn't work." By the time Claude is oriented, you've burned half your focus.
The fix is not a longer system prompt. The fix is a small set of markdown files, owned by you, loaded by Claude at the start of every session. Six files. One folder. Every one answers a question Claude would otherwise have to re-ask.
This guide shows you exactly what to put in each file, how to keep them in sync as the project evolves, and how to get Claude to update them for you between sessions. It's the same system we use across Agent-J, Aspectr, and the CDE website.
What's inside
Each file answers a specific question Claude has at the start of every session. Drop them in your repo, reference them in CLAUDE.md, and they load automatically.
TECH_STACK.md
Languages, frameworks, infrastructure, versions, external services. The ground truth so Claude stops guessing whether you use Next.js or plain Vite.
PARITY_CHECKLIST.md
Non-negotiables Claude needs to match every time: naming, file structure, testing conventions, accessibility rules. Pulled from every code review you've ever given.
DOWNLOAD_FLOWS.md
End-to-end flows for the critical paths in your product. The diagrams Claude needs when it touches anything that spans more than one file.
TESTING_PROTOCOL.md
Exact commands, browser matrix, acceptance gates. So Claude finishes a task the way you'd finish it, not the way it thinks "done" looks.
SESSION_CLOSEOUT.md
A tiny template Claude fills in before you close the window. Decisions made, dead ends hit, next steps. Loaded first in the next session.
INVESTIGATIONS.md
A running log of approaches that failed and why. Saves Claude (and you) from re-running the same dead-end experiment three weeks later.
The full system: what goes in each file, how to wire it into CLAUDE.md, and the prompt that keeps it updated. One PDF, no gate.
Want to see this in action? Join Agent-J+ — every build session uses it.