claude-memory

Read and write structured project knowledge in `.claude/memory/` across sessions.

1|Updated Apr 23, 2026
One-click install
npx skills add https://github.com/project-ax/ax-next --skill claude-memory-project-ax
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: claude-memory
Source: https://github.com/project-ax/ax-next/tree/main/.claude/skills/claude-memory
Command: npx skills add https://github.com/project-ax/ax-next --skill claude-memory-project-ax

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

It prevents repeated reinvestigation by persisting project facts, decisions, reusable patterns, and mistake lessons across sessions, while also capturing how you personally work so future sessions improve.

Core Features & Use Cases

  • Session read/write loop for durable working memory: Reads .claude/memory/ at session start, captures new corrections mid-session, and updates memory at session end when meaningful changes occurred.
  • Five structured memory artifacts: Maintains context.md, decisions.md, patterns.md, mistakes.md, and meta.md with specific purposes and rules (no padding, date-stamped entries, mistake graduation).
  • Safety for parallel agents: Ensures memory writes and commits happen in your own worktree/branch copy to avoid silent decision loss under concurrent runs.

Quick Start

Use the claude-memory skill at the start of a meaningful project task, then when you finish say “done” (or “wrap up” / open a PR) so the skill updates .claude/memory/ based on what changed.

Frequently Asked Questions about claude-memory

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I persist project memory and context across coding sessions?

You can persist project memory across coding sessions by reading and writing structured knowledge artifacts to a `.claude/memory/` directory at session boundaries. This maintains durable facts, decisions, and patterns to prevent repeated reinvestigation.

What is the best way to track mistakes and lessons learned in a codebase?

Tracking mistakes in a codebase is best handled by logging them in a structured `mistakes.md` file using date-stamped entries. The system applies mistake graduation logic to ensure resolved errors evolve into reusable patterns.

How do I maintain a decision log for ongoing software engineering work?

Maintaining a decision log for software engineering work involves writing architectural choices to a structured `decisions.md` file within your repository. This ensures session-to-session continuity for project rationale without padding.

Can I use project memory files safely when running parallel agents?

Project memory files are safe for parallel agents when memory writes and commits target your own worktree or branch copy. This isolation prevents silent decision loss and merge conflicts under concurrent runs.

What files do I need for repository-level working memory?

Repository-level working memory requires five structured markdown artifacts: `context.md`, `decisions.md`, `patterns.md`, `mistakes.md`, and `meta.md`. Each serves a specific purpose for capturing project know-how and self-improvement observations.

When should I update session continuity files during a task?

You should update session continuity files at meaningful task milestones: session start, mid-session when corrections occur, and task completion. Triggering a memory update by saying "done" or "wrap up" captures what changed.