Spec v0.1

The package standard for agent-operable work.

Workflowware is software packaged as an agent-operable workflow — readable by humans and agents, with the job, rules, evals, and handoff in one kit.

View SPEC.md on GitHub → Open package template

Definition

What a package must do.

Explain the job, define the workflow, describe agent behavior, state safety boundaries, include evaluation examples, and provide handoff metadata.

Minimum package

MVP file set.

workflowware-package/
  README.md
  index.html
  workflow-spec.md
  agent-spec.md
  install-guide.md
  approval-rules.md
  eval-template.csv
  agent-handoff.md
  artifact-manifest.json

Required layers

Concepts every kit carries.

Human trust surface

What it does, why it exists, how it’s controlled, and what proof exists.

Workflow spec

Trigger, context, old/new workflow, finish line, edge cases, success metrics.

Agent spec

Role, inputs, tools, allowed/prohibited actions, outputs, escalation.

Governance

Approvals, safety boundaries, audit trail, human review, non-autonomous zones.

Evals

Example inputs and expected behavior you can test and improve.

Handoff + manifest

Install path, customization notes, and machine-readable package map.

Manifest

Machine-readable fields v0.1.

{
  "title": "string",
  "artifact_type": "workflowware_package | …",
  "status": "draft | prototype | active | published",
  "primary_artifact": "file:///… or https://…",
  "summary": "string",
  "files": [],
  "related_artifacts": [],
  "agent_handoff": "path",
  "registered_in_maat_memory": "uuid or pending"
}

Full schema: MANIFEST.schema.json

Next

See a live package.

Missed Lead Recovery is the first public pilot built on this spec.

Open the pilot → Browse the library