Methodology
HELIX
Frame the work with specs, design decisions, and test plans. Use the alignment skill to keep those documents in sync as the project changes.
Open-source developer tools
Specify what to build, share schemas across tools, and show readers how published work was made.
Methodology
Frame the work with specs, design decisions, and test plans. Use the alignment skill to keep those documents in sync as the project changes.
Declarative design
UMF is our foundation for declarative system design. The UMF family builds on shared, machine-readable models for data and actions.
Provenance
Publish a brief colophon that declares how a piece was made. Sign it when you want readers to be able to verify the declaration.
Declarative system design
Declare records, fields, relationships, and actions in UMF. Choose how to store the records, query them, execute actions, and expose them to clients.
A worked example
Design a CSV export from the acceptance criterion through data, storage, actions, and an API. This example illustrates how the pieces fit together.
Write an acceptance criterion a person or an agent can test.
When a date filter is active, the CSV contains only the visible reporting period and starts with column headers.
Declare a ReportRow with a reporting date, category, and decimal total. Model an ExportJob with a date range, state, and a reference to the report.
Each CSV row must contain a reporting date, a category, and an exact decimal total.
Track the export job and its link to the report with Truss on PostgreSQL. Read report rows from a published snapshot on Ashlar in Databricks.
Record the snapshot ID with the job so a retry can use the same reporting data.
Use generated Python access to look up a job by key on a supported profile. Run native SQL for a backend-specific read, or use Weft to compile the report’s date-range query into SQL for the chosen backend.
Pass the date range as query parameters and stream the resulting date, category, and total into the CSV.
Declare actions to request the export and mark it ready. The host coordinates the saga across the job store and file storage.
Reserve a job ID, stage the CSV, then mark the job ready. Retry with the same ID. If publication is abandoned, remove the staged file and mark the job failed.
Both requests look up export-42 and return its state and reporting date range. These sample routes and fields are illustrative.
POST /graphqlquery ExportJobStatus {
exportJob(id: "export-42") {
id
state
startDate
endDate
}
}
{
"data": {
"exportJob": {
"id": "export-42",
"state": "READY",
"startDate": "2026-10-01",
"endDate": "2026-10-31"
}
}
}
GET /export-jobs/export-42
Accept: application/vnd.api+json
{
"data": {
"type": "export-jobs",
"id": "export-42",
"attributes": {
"state": "READY",
"startDate": "2026-10-01",
"endDate": "2026-10-31"
}
}
}
The application authorizes access to the job and serves the CSV only after the job is ready.
Attach a short declaration to the release note, with an optional signature for verification.
Mixed authorship. An AI assistant drafted the note. The maintainer checked it against the shipped feature and approved the text.