Your AI agent forgets everything you tell it.
It crashes with no recovery. Burns through context in 10 minutes. Makes the same mistakes over and over. And when you give it something complex, it tries to do everything in one thread until it collapses.
Sound familiar?
I got sick of it. So I built 5 skills that fix all of it. They're free, they need zero API keys, and they work with any OpenClaw agent out of the box.
Here's what they do and why you need them.
01 — Smart Memory
The problem: Your agent forgets every conversation the second it restarts. You tell it your preferences on Monday, by Tuesday it's asking you again.
What this does: Auto-saves every decision, preference, and lesson to actual files. Daily notes get consolidated into long-term memory automatically. Your agent builds a persistent knowledge base about you, your projects, and your stack — without you lifting a finger.
No database. No embeddings API. No OpenAI key. Just files that survive restarts.
02 — Self-Corrector
The problem: Your agent makes something up, you correct it, and it says "sorry, you're right!" Then makes the exact same mistake tomorrow.
What this does: When you correct your agent, it doesn't just apologise — it updates its own rules. It writes the correction to a learnings log, then promotes important patterns to your workspace files (AGENTS.md, TOOLS.md) so it literally never makes the same mistake twice.
It also catches confabulation before it reaches you. If your agent is about to claim "73% of developers prefer..." without a source, the pre-send check flags it.
03 — Context Budgeter
This is the one nobody talks about.
The problem: Your agent silently burns through its context window. You're 45 minutes into a complex task, it hits the limit, compacts, and loses everything you were working on. No warning. No recovery.
What this does: Monitors context usage in real-time. Warns you at 50%, 70%, and 85%. When it gets heavy, it auto-spawns sub-agents to handle the remaining work instead of burning your main context. Before any compaction, it writes checkpoints so your working state is preserved.
You never lose work to compaction again.
04 — Error Recovery
The problem: Gateway crashes at 2am. Browser session dies. Cron scheduler breaks. You wake up to a dead agent and have to debug it manually.
What this does: Detects failures automatically. Gateway unresponsive? It restarts it. Browser dead? It recovers the session. Cron broken? It fixes the scheduler. All with exponential backoff — try, wait, retry, escalate.
It only pings you after 3 failed recovery attempts. Everything else is handled silently.
Comes with a standalone health check script you can run from your heartbeat or cron.
05 — Task Router (my favourite)
The problem: You give your agent a complex request and it tries to do everything in one thread. Reading files, calling APIs, writing code, running tests — all in the same context until it drowns.
What this does: Scores every incoming request on complexity (0-10). Simple stuff runs inline. Complex stuff auto-spawns a sub-agent. Really big tasks get decomposed into parallel tracks — frontend agent, backend agent, deploy agent, all working simultaneously.
It also includes model selection guidance. Simple tasks use cheap models. Complex tasks get the expensive ones. Your token spend drops without you thinking about it.
This is the one that turns a chatbot into an orchestrator.
The bottom line
Same model. Same setup. Completely different agent.
The difference is these 5 installs. They make your agent fundamentally better at being an agent — better memory, better learning, better resource management, better resilience, better task handling.
All free. Zero API keys. Zero paid dependencies.
I built them because nothing like this existed. Every "agent skill" I found on the market was either a thin prompt wrapper or required expensive API subscriptions. These actually work.
Quick links
Skill | What it does | Link |
|---|---|---|
Smart Memory | Never forgets | |
Self-Corrector | Learns from mistakes | |
Context Budgeter | Manages context | |
Error Recovery | Self-healing | |
Task Router | Smart orchestration |
If you found this useful, forward it to someone building with AI agents. They'll thank you.
And if you want to see these in action, I made a reel showing the before and after — check my Instagram @mikemeansbusiness_ai.
Until next time,
Mike