bookend

name	bookend
description	This skill should be used when the user asks to "set up bookend", "morning briefing", "session start", "session end", "checkpoint", "save state", "capture context", "evening wrap", "end of day", or discusses memory management, context persistence, or session continuity for AI assistants.
version	1.0.0

Bookend — Anti-Context-Loss System

Never lose context again. Bookend gives your agent a structured memory system with morning briefings, checkpoints, and end-of-day routines.

The Problem

Your agent forgets everything when context compacts. You say "do point 5" and it has no idea what you mean.

The Solution

Bookend creates a state/current.md that survives any compaction. Your agent reads it on startup and always knows what's going on.

Features

• Morning Briefing — Agent wakes up, reads state, checks emails, briefs you
• Auto-Checkpoints — Saves progress every 30 min and after every task
• End of Day — Finalizes state, updates daily memory
• Nightly Builds — Agent works on a backlog while you sleep
• Survives Compaction — state/current.md is the single source of truth

Setup

Create the following files in the user's workspace:

1. state/current.md — Live state file with sections: Active Threads, Recent Context, Pending Tasks, Notes
2. state/ROUTINES.md — Morning/checkpoint/EOD routines
3. state/nightly-backlog.md — Tasks for overnight builds
4. memory/ directory — For daily memory files
5. MEMORY.md — Long-term curated memory
6. Update AGENTS.md with the bookend checkpoint rules

How It Works

Morning (first heartbeat after 07:00)

1. Read state/current.md
2. Check yesterday's memory file
3. Scan emails for urgent items
4. Send brief: Top 3 priorities, new emails, next appointment

During the Day (every ~30 min)

1. Update state/current.md with current topics
2. Append to memory/YYYY-MM-DD.md
3. If context is getting full, checkpoint NOW

End of Day

1. Finalize state/current.md
2. Close daily memory file
3. Review if anything belongs in long-term memory

Overnight (optional cron at 23:00)

1. Read state/nightly-backlog.md
2. Work on top item
3. Document in memory/nightly-builds.md
4. Don't disturb the human

Why This Works

The key insight: your agent needs a "current state" file that's always up to date. When compaction happens, the agent reads this file and picks up exactly where it left off.

• 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