Stop re-teaching Claude every session.
Give Claude persistent memory, QA gates, and a learning loop that compounds — so the same mistake doesn't cost you twice.
Built for solo operators and small teams who are tired of Claude forgetting lessons between sessions.
The Problem
Claude forgets every lesson.
You teach it once. Next session, the same mistake. No context carryover.
Ships break on a clean machine.
Works in your Cowork tab. Deploy it, and the gap shows up too late.
No QA gates before shipping.
Claude finishes a task, you call it done. No pre-flight check, no proof it actually ran end to end.
Every team starts from zero.
No playbook. No patterns. No way to share what worked with other operators.
“Claude shouldn't forget how your system works.
— the reason this kit exists
The Solution
Persistent Lesson Database.
Every error and every win becomes a lesson Claude can load on the next session. The point is to make repeating the same mistake far less likely.
Pre-Release Quality Gate.
Six checks before anything ships. Fresh-install test, runtime validation, schema and constraint scans. Catches the most common pre-ship failures.
Reinforced Learning Loop.
Every approach gets logged with what worked, what did not, and why. Reinforce the good patterns, retire the bad ones, share the playbook across your team.
Seeded Starter Lessons.
Do not start from zero. Comes with battle-tested lessons covering Windows filesystem gotchas, Cowork stale reads, retry patterns, and more.
Features
- Schema v2 Lesson Database. Type-safe SQLite for storing patterns, fixes, and generalizations.
- Six-Gate Ship Check. Fresh-install runtime, schema validation, constraint audit, integration test, deployment safety, evidence standard.
- Fresh-Install Runtime Oracle. Runs your code in a clean tempdir before you call something done.
- 30+ Seeded Lessons. Starter pack covering Cowork gotchas, Windows filesystem access, API retries, and more.
- Legal Pack. Compliance templates, evidence standards, audit checklists.
- Idempotent Install. Safe to re-run on setup changes. Same result every time.
- Portable Installs. COWORK_BRAIN_KIT_DIR lets teams point at shared lesson DBs.
- Cross-Platform. Windows, Mac, Linux.
Requirements
Python 3.8+
Cowork mode
Around 5 minutes to install
No code changes required
Why this works when prompts don't
Claude doesn't persist state.
Every new session resets. Prompts aren't memory — they're re-explanations.
Memory lives outside the model.
A structured SQLite lesson DB Claude can load, search, and write back to. Externalized, not ephemeral.
Quality gates run before you ship.
Six checks — fresh-install runtime, schema validation, constraint audit — catch issues in seconds, not in production.
Common questions
“Isn't this just prompts and templates?”
No. It's a structured SQLite lesson database plus enforced runtime checks. Prompts vanish at session end. Lessons compound across sessions.
“Can't I build this myself?”
You could. This saves you roughly 10–15 hours of scaffolding and ships with 30+ battle-tested lessons covering the gotchas you'd hit anyway — Windows filesystem drift, stale FUSE reads, retry loops, and more.
“Will it work with my setup?”
Works with Claude Code and Anthropic Cowork. Python 3.8+. No code changes required. Set one env var and Claude uses your lesson DB automatically.
“What does it actually catch?”
Missing env vars, schema drift, broken imports, stale file reads, fresh-install failures, migration order bugs, silent API retries. The specific classes of failure that slip past “it works on my machine.”
Want Claude to actually remember what it learned?
v2.0.0 is ready. Persistent memory, quality gates, and a reinforced lesson library in about five minutes.