#Develop State and formal Storage

Define each Stage's business Schema and initializer. The host validates initializer input and each State write.

#Update complete State

Parse access.state.value with your Schema, construct a replacement JSON value, and await state.update. create requires an absent value; clear removes it without automatic reinitialization. Read views must not be mutated.

Cross-stage context should be initialized from deliberately saved Artifacts/Records and the actual Run identity.

#Commit formal data

Only Code receives writable Storage. For example, create one Artifact and append one Record atomically:

await access.storage.commit([
  {
    type: "create_artifact",
    id: "draft-1",
    payload: {
      loomName: "review",
      stageName: "review",
      data: { approved: true },
    },
  },
  {
    type: "append_record",
    id: "review-1",
    payload: {
      loomName: "review",
      stageName: "review",
      data: { decision: "approved" },
    },
  },
]);

This is a Code-body fragment. Choose unique IDs for real execution; these fixed IDs cannot be reused for repeated creation.

Replacing/removing an Artifact requires its observed expectedRevision. A batch rejects duplicate category/ID targets. Record payloads remain immutable, though Code can explicitly remove a Record.

Commit atomicity covers supplied Storage operations, not State updates, file changes, model calls or subsequent steps. Keep prior successful commit evidence after failure rather than assuming rollback.

See SDK operations and data concepts.