One fullstack application: a place to capture and revisit fieldnotes. Its shape and contracts come first.
Composition + provenance
The Astro starter and component system are composed as Parts. Source S1, output B1 and runtime R1 remain distinct identities: the build comes from the source, and activity belongs to that runtime.
Three views of the same System. Choose a view; read its description below.
Genome · The shape
Give the application its shape.
Fieldnotes is an illustrative application for capturing and revisiting observations. Start with the people it serves, what it should do and what must stay true. Screens, routes and private data take a recognizable shape before they need implementation.
A Thing can begin with its Genome alone. Parts and source are optional.
Proteome · The application
Make that shape tangible.
An Astro starter and a component system contribute to one application. Its screen, API and storage become solid, while the System keeps track of how they were realized and checked.
The recipe explicitly replaces the starter Button and adapts callers and configuration. Authored source and each realized output keep separate identities.
Phenome · The activity
See what happens when it functions.
Events and metrics sit beside the application they describe. Trace a saved record, inspect a request and see how the service behaves, with the source and output that produced it.
Activity here is illustrative. Whether it helps people capture and revisit useful observations remains unmeasured; functioning alone does not establish that purpose.
Start with a purpose. Keep it in the work.
An agent needs to know why a change matters, whose needs it serves, and what must stay true.
Artyfact gives that context a home: a System, the context and governance wrapper around one or more Things. A single Thing can be a whole application and begin with its Genome alone.
Keep in Touch, the separate example on our model page, explores another purpose. The purpose is more wanted contact with people you choose. Private notes, consent and a calm experience follow from that purpose.
A clear handoff. Room for different approaches.
Use your own agent and local tools. Artyfact supplies the task, context and isolated workspace.
1
Choose an Evolution
Name the change you want, the subjects it affects and the criteria it must meet.
2
Start a Realization
Get a scoped task packet and Git fork. Run a compatible agent locally; several approaches can proceed in parallel.
3
Submit a candidate
Checkpoint, push and submit an exact version. Inspect checks, evidence and overlaps with other work.
4
Review and promote
Review the candidate against the accepted intent. Promotion records the accepted System revision; deployment is a separate action.
Keep the distinctions that matter.
Agents you choose
The minimum workflow uses your local agent harness. You choose the provider and orchestration.
Parts you can reuse
Optional Parts can be partial or complete at any scale. Composition and upgrades are explicit, saved results with provenance.
Evidence you can inspect
A check has an execution state, a verdict and provenance. Unknown stays unknown.
Source and output apart
Authored source and each realized build output keep separate identities, linked through the Thing and Genome.
Bring your agent. Bring your intent.
Start with how the handoff works. Explore the model when you want the deeper picture.