Agent Memory Kit

Your agent forgot again.

You told it something important three sessions ago. It was on board. It agreed. Now it's suggesting the exact thing you ruled out — and you're back to explaining context you've already explained twice.

This is the OpenClaw memory problem. Sessions reset. Compaction eats details. The agent wakes up fresh every time, and "fresh" means amnesiac.

The Agent Memory Kit fixes this.

---

What changes after you install it

Your agent writes things down the moment they happen — not "later," not after you remind it. Decisions, commitments, facts, corrections, positions taken — all go to files in the same turn they appear. If it's not in a file, the kit treats it like it never happened.

Between sessions, a STATE.md file tells your agent exactly where it left off: active tasks, what's off limits, what's waiting on you. It wakes up oriented, not blank.

During long sessions, checkpoints fire every ~45 minutes so compaction doesn't silently kill your context. Before a session ends, a final flush captures everything worth keeping. A nightly cron handles memory hygiene so things don't pile up into noise.

The result: an agent that carries its history forward instead of starting from zero every day.

---

What's in the kit

Six files. No dependencies. No configuration beyond dropping them in your workspace.

• MEMORY-KIT.md — The core memory architecture. This is what your agent reads to understand how memory works and why immediacy matters.
• STATE-template.md — Cross-session state tracker. Includes a HARD CONSTRAINTS block so your agent knows what's ruled out before it starts suggesting things.
• AGENTS-memory-block.md — A ready-to-paste section for your AGENTS.md that embeds the write-first protocol directly into your agent's startup behavior.
• memory-hygiene.cron — Nightly maintenance job. Keeps memory files clean and useful instead of bloated.
• README.md — Five-minute setup. Literally five minutes.
• EXAMPLES.md — Before/after failure scenarios. Real cases showing the exact mistakes this prevents and how the kit would have caught them.

---

Already have lossless-claw?

Good. Keep it. Lossless-claw handles your conversation history — it's excellent at what it does. But it doesn't teach your agent to write decisions to files, track task state across hard resets, or maintain curated long-term knowledge. Those are different problems.

Here's the boundary: lossless-claw preserves what was said. The Memory Kit persists what matters. Your task state, behavioral commitments, preferences, active constraints — those live in files, not in a conversation database. A /reset wipes the lossless-claw DB. STATE.md and MEMORY.md are still on disk.

The ideal setup is both layers together. This kit is the behavioral layer lossless-claw can't replace.

---

For power users who've tried other memory solutions

Most memory products for OC patch one thing. They give you a MEMORY.md template, or a daily log format, or a reminder to "write stuff down." That's not enough, because the problem isn't knowing you should write things down — it's that agents don't do it in time. By the time the cue fires, the context is already gone or compacted away.

This kit is built around a single principle: write now, not later. The architecture creates specific triggers for the moments when memory loss actually happens — and handles the write in that same turn. That's the gap other solutions miss.

---

Setup

1. Copy the 6 files into your OC workspace
2. Paste the AGENTS-memory-block.md content into your AGENTS.md
3. Add the cron job
4. Done

Your agent reads the kit on startup and applies it immediately. No prompting required.

---

$9.

Your agent remembers. Your context survives sessions. You stop re-explaining the same things.

That's the trade.