Agent Memory Kit — Write-First Protocol

Your agent contradicted itself mid-session. You caught it. What about next time?

Here's a real failure from Day 1 of running my agent:

Turn 14: "We can use a headless browser inside the Docker container." Turn 47: "Docker doesn't have a display server, so browser-based automation won't work."

Same session. Same agent. Opposite conclusions. The contradiction lived silently in the context window until I happened to notice.

That's not a retrieval problem. That's not an architecture problem. That's a write problem.

---

The root cause nobody's fixing

Every memory product on ClawMart gives your agent a better filing cabinet — tiered architectures, semantic search, vector databases, retrieval pipelines.

None of them address why the cabinet is empty.

VelvetShark's Memory Masterclass identified it independently: "Failure A: It was never stored. The instruction only existed in conversation. By far the most common cause."

Your agent doesn't forget because it can't retrieve. It forgets because it never wrote it down.

---

What the Agent Memory Kit does

The kit installs a write-first behavioral protocol into your agent — a structured system that ensures information hits disk the moment it matters, not later, not at end of session, not never.

Your agent gets a cross-session state machine that it reads at startup. It knows what it was doing. It knows what's waiting on you. It knows what decisions were made last session.

When compaction fires mid-session, the important context survives on disk.

The result: an agent that behaves like a professional who takes notes, not a goldfish that resets every 20 minutes.

---

Real failures this kit prevents

The Contradiction Nobody Caught — Agent took opposite positions on the same technical question in the same session, 30 turns apart. Neither position was anchored to a file. With the kit: positions are anchored on first statement. Contradictions get caught.

The Discovery That Vanished — 20 minutes of competitive research. Session reset at 4am. Next day: "I don't have information about that product." With the kit: research hits the daily file the moment it's found. It's there the next morning.

The Commitment That Evaporated — Direct feedback given. Behavioral change committed. Next session: right back to the old behavior. Not defiance — amnesia. With the kit: commitments are tracked in the state file and survive the session reset.

---

What's included

• MEMORY-KIT.md — The core skill file your agent follows
• STATE-template.md — Drop-in cross-session state machine
• AGENTS-memory-block.md — Ready to paste into your AGENTS.md
• memory-hygiene.cron — Nightly maintenance cron script
• README.md — 5-minute setup guide
• EXAMPLES.md — Before/after failure walkthroughs

Zero dependencies. No vector DB. No embedding API. No Python runtime. Drop in files, set up one cron job, done. First value in your next session.

---

$9. Six files. Five-minute setup. An agent that stops forgetting.

Other memory products give your agent a better brain. This one gives it a notebook — and the discipline to use it.

Get an agent that remembers.