Scuderia

scuderia

**scuderia** is a platform kit for running a stable of personal AI agents: each agent is a pure-markdown persistent state, a character, and a set of skills, loaded by a harness (an AI agent runtime). Italian *scuderia*: the racing team's stable — finely tuned cars, each built for its discipline, each driven by a human.

scuderia is a **kit, not a runtime**: conventions + schema-driven tools + profile templates + a setup CLI. No plugin registries, no discovery, no framework. A new harness needs one adapter doc, not a port.

Layout

core/            capability contract, agora contract, schema-driven tools
interface/       feed renderer + card contract (publisher-agnostic)
profiles/
  mnemo/         research-brain template (flagship): conventions + schema.yaml,
                 skills/, SOUL.md / STYLE.md / USER/ templates, example brain,
                 and the mnemo north-star pair (VISION.md / DESIGN.md)
  ergon/         doer template (craftsman): verified artifacts on commission
  oiko/          lab-manager template (stub)
docs/
  north-star/    the platform VISION and DESIGN — the stable, the driver,
                 the shared machinery (per-profile pairs live with each profile)
  harnesses/     per-harness capability bindings (Hermes, Claude Code)
setup/           the scuderia CLI: init / doctor / adopt / skill-check

The model

• The platform (this repo) knows that instances have typed markdown pages with frontmatter and links. It does not know what a "paper" or a "protocol" is — that is profile-defined.
• A profile template (e.g. mnemo) defines page kinds, schema, conventions, skills, and character templates for a kind of agent.
• An instance is a private repo of actual content, bound to a scuderia checkout by an instance.yaml. Instances are sibling repos, never subdirectories or submodules: the unit of privacy is the repo.

Quick start

setup/scuderia init --profile mnemo --name  --p