Your project knowledge compounds and learns across sessions.
Claude Code is stateless. Every session starts from zero — no memory of yesterday's decisions, no record of bugs already fixed, no knowledge of what you tried and rejected last week. You re-explain. Claude re-suggests the same things. The same mistake happens twice.
Clankbrain fixes that. It gives Claude Code a living memory that grows with your project — capturing decisions, lessons, and patterns every session, so each session starts smarter than the last.
- You use Claude Code daily on a real, ongoing project
- You've already felt the pain of re-explaining your codebase every session
- You're disciplined enough to run two commands:
Start SessionandEnd Session
If you're just experimenting with Claude Code, come back when it's your primary tool.
After 8 sessions, type Progress Report:
=== Clankbrain Progress Report ===
Sessions logged 8
Lessons accumulated 14
Known errors logged 6 <- never debugged twice
Rejected approaches 9 <- never re-proposed
Skill accuracy 78%
Last 3 sessions:
[2025-05-12] Email throttle + scheduler fix (4 file saves)
[2025-05-14] Dashboard accordion redesign (7 file saves)
[2025-05-15] Fixed IDENTITY column error (2 file saves)
-> 8 sessions in. Compounding is happening.
Session 1 starts blank. By session 8, Claude knows your patterns, your mistakes, and your decisions — and applies them automatically.
npx clankbrainSetup auto-detects your project name and tech stack. No API keys. No background service. No database. Python 3.7+ required.
Requires: Claude Code
Start Session -> reads memory, applies past lessons, picks up where you left off
[work]
End Session -> extracts lessons, saves everything to memory
Two commands. Everything else runs automatically.
- Persistent memory — decisions, bugs fixed, rejected approaches, codebase knowledge
- Skills that self-improve — each skill scores itself;
/evolvepatches the ones that keep failing - Drift detection — catches undocumented changes after every file edit
- Regret guard — scans past rejected approaches before every prompt, blocks re-proposing them
- Progress reports — real numbers built from your actual session history
Clankbrain compounds with use — but only if you use it. A developer who runs Start Session / End Session every session and /evolve every few weeks will have a Claude that gets measurably better at their specific codebase over time. Someone who uses it sporadically gets marginal gains.
Tested across 140 real sessions on a production codebase. Not a demo project.
- Skills and the learning loop
- Lifecycle hooks
- Every command
- Architecture, modes, and file tree
- Cross-machine sync and team sync
- Other IDEs and install options — Cursor, Windsurf, Warp, GitHub Copilot
- FAQ
Built by Yehuda Frankel. Using it on a real project? Tell us what you're building -> - If it helped, star it — it helps others find the original.
