<aside>
🧠 Introduction
You have an agent that works for four days and then starts forgetting what it knew, citing books to look smart, and saving files where you'll never find them again.
You've rewritten the system prompt six times. Every rewrite bought you another few days.
The prompt was never the problem. You built an agent with no identity, no searchable memory, and no rule about what it's allowed to do without asking.
This kit is the fix. Three templates you can fill in tonight - the Constitution, the autonomy matrix, and the file routing table - plus the ten-minute starter that gets you 80% of the foundation before the weekend.
You don't need to be an engineer. You need a text editor, one folder, and the willingness to write down what you actually want.
Read time: 12 minutes. Setup time: one afternoon. Time to first payoff: your next session.
</aside>
<aside>
I spent six weeks building a personal agent. It managed tasks, tracked deals, read email, queried raw business data, and surfaced things I'd otherwise miss.
Week one it worked for ten minutes at a time. Then it hit a situation my prompt didn't cover, invented a policy, and applied it confidently. So I made the prompt bigger. Added rules. Added the rules under the rules that kept getting ignored.
Every fix made it worse. The agent wasn't confused about the rules - it had no foundation the rules could sit on.
I stopped writing prompts and wrote a Constitution instead. Then a memory system split by domain. Then an autonomy matrix built on what could be undone. It worked on the first run after that.
This is the document I wish someone had handed me on day one - the three files that carry the whole system, filled in, not bracketed.
</aside>
<aside>
✅ Before You Start - Quick Checklist
<aside>
1. The Ten-Minute Starter: The IDENTITY.md file that gets you 80% of the foundation before you do anything else.
2. The Constitution Template: Filled in, not bracketed. Identity, principal, hard rules, guidelines, what-it-is-NOT, THINK vs DO.
3. The Constitution Interview Prompt: The prompt that refuses to write anything until it has interviewed you properly.
4. The Autonomy Matrix: L0 through L4, each level tagged reversible or irreversible, with the confirmation rule for each.
5. The File Routing Table: Where every file type lives, so the agent stops inventing seven homes for the same output.
6. The Memory Folder Map: The domain-split structure and the MEMORY.md index that keeps memory searchable past week two.
7. The Weekly Build Order: All the work sorted into five weeks, in dependency order, so you build the foundation before the amplifiers.
8. The Pre-Flight Checklist: What to verify before your agent takes any action unsupervised.
</aside>
<aside>
If you already run a versioned, spec-driven agent in production, you know this. This was built for everyone else.
</aside>
<aside>
Step 1 - Duplicate this Notion page: Hit the ••• menu top right → Duplicate.
Step 2 - Do the Ten-Minute Starter now: Four things into one file. That's most of the foundation. Everything after is refinement.
Step 3 - Run the Constitution Interview: Section 3 is a prompt. Paste it into your agent and answer honestly. It writes the real Constitution from your answers.
Step 4 - Fill the three tables: Autonomy matrix, routing table, memory map. Replace the examples with your real actions, folders, and files.
Step 5 - Follow the build order: Section 7 sorts everything into five weeks. Build in dependency order, not in the order that feels fun.
🔰 New to this? A "personal agent" just means an LLM wired to your own files and rules, instead of a fresh chat every time. That's the whole concept. No framework required - it's a folder, some markdown, and a document that tells the agent who it is.
</aside>
Log each piece as you finish it, and where the agent still breaks. After a week, you'll see your own weak spot.
Build Tracker - The Personal Agent Build Kit
| Date | What I built | Category | Where it still breaks | Fix |
|---|---|---|---|---|