LYN_0 Academy

Build your own personal AI agent, one mission at a time.

Most people use an AI that starts every conversation at zero. It doesn't know your name, your
job, your files, or what you told it last Tuesday. An agent is different: it has an identity
and a memory of its owner, both loaded before it ever sees your message.

This repository holds the mission packs. One folder per week. Install it, run it, and your
agent becomes measurably less generic every week.

Mission 01 — Your AI learns who it's working for

Status: free and complete. This is the whole pack, nothing held back.

Time: about 45 minutes, once.

Quick start

cd 01-know-your-owner
python install.py            # installs SOUL.md and memories/USER.md
# now edit both files and replace every ___ with your own answers
python install.py --check    # tells you what's still blank or missing

Then open test.md and run the six checks in a new chat.

install.py --force overwrites without asking. Existing files are backed up to
<hermes-home>/backups/ automatically, with a timestamp, every time.

Requirements

• Hermes Agent installed and running on your own
  machine.
• Python 3.9 or newer for the installer. Nothing else. No dependencies, no network calls.

On Windows the Hermes home is %LOCALAPPDATA%\hermes, on macOS and Linux it's ~/.hermes.
install.py resolves it for you, or pass --home to point somewhere else.

Mission 02 — Memory that survives

Status: free and complete. Same deal, the whole pack.

Time: about 40 minutes, plus 5 minutes a week from now on.

Quick start

cd 02-memory-that-survives
python install.py            # starter MEMORY.md (only if you have none) + the review skill
python install.py --audit    # every entry, its size, the duplicates, the total against budget
python install.py --check    # what's still missing, changes nothing

--audit on a real setup looks like this:

MEMORY.md: 9 entries, 1889/2200 chars (85%)
    [ 552] Project Hyperliquid trading (...)   <-- too long
    [ 332] LYN_0 TikTok project (...)          <-- too long
USER.md:   6 entries, 1205/1375 chars (87%)

Two entries eating a fifth of the budget each, and the file already over 80%. That's what memory
rot looks like before anyone notices: it doesn't look stale, it looks like a fact.

What this actually gives you

SOUL.md and USER.md are not documentation. They are injected into the system prompt on
every single turn. That means:

• what you write there is what your agent knows, forever, starting now
• an agent with a name stops feeling like a tool
• "never touch this folder" in the file is worth more than saying it again next week

The next missions add memory that survives, hands, documents, automations and a way to check
its own work. Mission 01 is the part everything else assumes.

Mission 02 is the part that makes it yours: it gets a memory file it writes itself, the rules
for deciding what goes in it, a weekly review to stop it rotting, and a backup so it survives
the laptop. All of it is plain markdown on your disk. If a file has it, your agent knows it; if
no file has it, it never knew it.

Roadmap

Missions 03 to 12 ship weekly. 01 and 02 stay free.

Feedback wanted

Run the six checks and tell me where it breaks. The failure I'm most curious about is the
agent contradicting the file you just wrote. Open an issue with what you wrote in SOUL.md
and USER.md (redact anything private) and what the agent did instead.

License

MIT. See LICENSE. Use it, fork it, ship it, sell whatever you build with it.