Never lose context again. Structured memory system for AI assistants.
# In Claude Code
/plugin install bookendOr clone manually:
git clone https://github.com/rockywuest/bookend-skill.git ~/.claude/plugins/cache/bookend/latest/setup
This creates state/current.md, memory/, MEMORY.md, and adds the bookend protocol to your AGENTS.md.
Your agent will now automatically:
- Run a morning briefing at session start
- Checkpoint every ~30 minutes
- Wrap up at end of day
Say "checkpoint" anytime to trigger a manual save.
AI assistants forget. Session ends, context gone. Even within a session, context compaction loses details. After a week of "learning" your preferences, your assistant might:
- Present old news as new
- Forget decisions you already made
- Repeat mistakes you corrected
- Lose important context mid-conversation
Bookend implements a "bookend" pattern — structured check-ins at session start and end that capture state and ensure continuity.
Morning Briefing (Session Start)
├── Read state files → know where things stand
├── Check memory files → recall recent context
├── Scan for pending tasks → nothing falls through cracks
└── Brief the human → 30 second catch-up
Evening Wrap (Session End)
├── Capture today's events → daily memory file
├── Update state files → where things stand NOW
├── Flag open threads → what needs follow-up
└── Update long-term memory → lessons learned
Your assistant wakes up fresh each session, but these files ARE its memory. Structured files beat vague "remember this" instructions every time.
- Read
state/current.md - Check yesterday's memory file
- Scan for urgent items
- Brief the human: top priorities, open threads, next steps
- Update
state/current.mdwith current topics - Append to
memory/YYYY-MM-DD.md - If context is getting full — checkpoint NOW
- Finalize
state/current.md - Close daily memory file
- Review if anything belongs in long-term memory
- Read
state/nightly-backlog.md - Work on top item from backlog
- Document in
memory/nightly-builds.md
your-workspace/
├── MEMORY.md # Long-term memory (curated wisdom)
├── AGENTS.md # Instructions (includes bookend protocol)
├── state/
│ └── current.md # Right now: open threads, recent context
└── memory/
├── 2026-02-15.md # Daily notes (raw events, decisions)
├── 2026-02-14.md
└── ...
- Explicit state — No "did we discuss this?" guessing
- Daily files — Easy to review, grep, reference
- Curated memory — Long-term file stays relevant (not just logs)
- Checkpoint discipline — Survive context compaction
/plugin install bookend
Then run /setup in your workspace.
git clone https://github.com/rockywuest/bookend-skill.gitTell your agent: "Read SKILL.md from the bookend skill, then set up the bookend system in my workspace."
Copy the snippet from templates/AGENTS-snippet.md into your AGENTS.md. No plugin needed.
| Directory | What's Inside |
|---|---|
templates/ |
Starter files for your workspace (state, memory, AGENTS.md snippet) |
examples/ |
Real-world examples of morning briefings, evening wraps, and filled state files |
references/ |
Patterns, anti-patterns, and best practices for using Bookend effectively |
Works with any AI assistant that can read/write files:
- Claude Code (native plugin support)
- Clawdbot / Moltbot
- Claude Projects (with file upload)
- GPT with Code Interpreter
- Cursor, Codeium, Windsurf
Built by Nox after experiencing 10+ memory failures in the first week of existence. The pattern is inspired by MARVIN and adapted for structured agent memory.
Lesson learned: Memory is not optional. It's infrastructure.
See CONTRIBUTING.md for how to contribute. This project follows the Contributor Covenant code of conduct.
See SECURITY.md for reporting vulnerabilities.
