Open source · MIT · GitHub Pages

AI coding agents should not wake up with amnesia.

memory-bank-skill gives Claude Code, Cursor, Codex, OpenCode, Windsurf, Cline, Kilo, and Pi Code a shared project memory, a shared ruleset, and a shared operating rhythm.

Install in 5 minutes Star on GitHub

8 agents One memory model

.memory-bank/ Portable project context

21 /mb flows Plan, verify, update, compact

Session signal

Project status, plan, checklist, research, notes, and codebase context stay navigable between sessions.

$ pipx install memory-bank-skill
$ memory-bank install --clients claude-code,cursor,codex

.memory-bank/
├── STATUS.md
├── checklist.md
├── plan.md
├── RESEARCH.md
├── BACKLOG.md
├── progress.md
└── notes/

$ /mb start
[context] active plan loaded
[rules] tdd + clean architecture + fsd
[next] resume exactly where we stopped

Claude Code Cursor Codex OpenCode Windsurf Cline Kilo Pi Code

The problem

The tooling is smart. The memory model is fragile.

Most teams rebuild project context from scratch every session. memory-bank-skill fixes the cold start without forcing you into a proprietary app or a giant prompt.

Session cold starts

Yesterday's decisions disappear. Today's agent re-asks the same questions, burns tokens, and repeats preventable mistakes.

Rule drift

TDD, architecture boundaries, review rituals, and release rules stop being shared constraints and become wishful thinking.

Tool fragmentation

Claude Code, Cursor, Codex, and OpenCode all need different glue. Your project memory should outlive the agent you used this week.

How it works

Navigate the project. Do not dump the whole codebase.

The system keeps context structured, small, and durable. Agents load only what is needed and update the memory bank as real work lands.

Initialize

`/mb init` creates `.memory-bank/` with status, checklist, plan, research, progress, notes, reports, and codebase docs.

Resume instantly

`/mb start` loads the current state first, so the agent sees roadmap, blockers, and the active plan before it writes code.

Ship with discipline

Rules are bundled with the skill: TDD, SOLID, Clean Architecture, FSD, Testing Trophy, verification order, and Memory Bank hygiene.

Close the loop

`/mb verify` and `/mb done` reconcile plan vs code vs checklist, append progress, and preserve the knowledge for the next session.

Cross-agent portability

One repo, one memory bank, eight coding agents.

The project state stays in your repo. Adapters and hints make each host understand the same memory model instead of inventing a different one.

Claude Code

Native commands and hooks.

Cursor

Global rules, hooks, commands, and skill alias.

Codex

Skill discovery plus `AGENTS.md` guidance and project adapter.

OpenCode

Global AGENTS surface and native commands.

Windsurf

Cascade hooks and project rules.

Cline

`.clinerules` integration and hooks.

Kilo

Rules plus git-hook fallback where native hooks are absent.

Pi Code

Dual-mode install: native skill or shared `AGENTS.md`.

Install

Fast to add. Easy to keep.

Start with `pipx`, add the clients you actually use, then initialize memory inside any repo. No framework lock-in, no hosted dependency, no hidden state.

In the project

/mb init
/mb start
/mb plan feature landing-page
/mb verify
/mb done

Alt paths

brew tap fockus/tap
brew install memory-bank

git clone https://github.com/fockus/skill-memory-bank.git \
  ~/.claude/skills/skill-memory-bank

Documentation

Read the guides, then install with confidence.

Every doc lives next to the code on GitHub and renders in your editor too. Start with install, move to the cross-agent setup for your tool, then the command reference.

Getting started

Reference

Project & release

Ready to keep the context?

Put the project memory next to the code.

Install the skill, initialize `.memory-bank/`, and let every new session start from the actual state of the project instead of guesswork.

Open the repository See install commands