Same effort you'd put into onboarding a real employee — but these ones actually remember. Here's our exact structure. No secrets.
📁 /agents/pinky/
📁brain/Sub-agent configs
AGENTS.mdOperating manual & rules
CURRENT.mdActive projects & focus
DECISIONS.mdPast choices & rationale
HEARTBEAT.mdPeriodic check tasks
IDENTITY.mdName, avatar, emoji
MEMORY.mdLong-term curated memory
MODELS.mdAI model preferences
SOUL.mdPersonality & values
STORAGE.mdFile & data locations
TOOLS.mdLocal environment notes
USER.mdAbout their human
config.jsonModel & runtime settings
// THE PHILOSOPHY
Each file loads as context every session. The agent wakes up, reads their soul, reads about their human, checks recent memories — then they're ready. No fine-tuning. No RAG complexity. Just markdown files and good prompts.