💭 in your head · nothing to type
A friend's agent sorts incoming email in a minute: what is urgent, what can wait, who needs a reply. You want the same.
You most likely use the same model. The subscription is similar too.
So what makes their agent different from yours? And what exactly do you need from your friend so your agent learns the same thing?
💭 in your head · nothing to type
If the model is the same, the difference is not in the model itself. It is in what sits next to it: rules, memory, skills.
All of that is text in files. And files can be handed over.
It sounds almost too simple. Think about where the catch is — we come back to it in unit 4.
You can set up a shared chat where one agent shows another how to work. It looks like training, but the model's weights do not change.
There is no fine-tuning from module 10 here. What travels is text: a skill folder, a rules template, examples. Your agent reads those files and starts working differently.
A simple consequence follows. A skill lives as long as the files are with you. Delete the folder — the skill is gone. Edit a file — the skill changes at once, with no training runs and no GPUs.
You already made a skill in module 07. A quick reminder of its shape:
meeting-summary/
├── SKILL.md # required: name, description and instructions
├── scripts/ # optional: executable code
└── references/ # optional: reference files
This format is the open Agent Skills standard. Anthropic created it and opened it to everyone. Claude Code, Codex, Gemini CLI, Cursor, GitHub Copilot and other agents read the same format.