Each kit contains the files that cultivate an agent, already structured and with a fictional example filled in so you can see the right size. Rules: one good page is worth more than twenty; every correction goes in the file, not in the conversation; weekly review; no secrets inside the files.
For chat AI (ChatGPT, Claude, Gemini). About me, rules, examples, memory, failure log. You paste it into the chat's instructions.
For an agent with tools (Claude Code, Codex, OpenClaw). CLAUDE.md and AGENTS.md, memory, failure log, security, two runbooks. Goes in the folder where the agent works.
For an agent in a process. Agent card, context, twenty annotated examples, scorecard, failure log. One process, one agent, one metric.
Kit files are in Portuguese; the structure and the LEIA-ME explain each file (an English version is in kits/LEIA-ME.en.md).
Would you rather have the agent write the files for you? Install the /cultivar and /revisao-semanal skills. Not sure where to start? Take the diagnostic. Want the card filled in right now? Use the generator. The repository is a GitHub template: "Use this template" creates your copy with everything.