Final Reference Flow¶

Target

This page describes intended behaviour or structure. No implementation is claimed unless a section is explicitly marked Current implementation or Evidence.

Reviewed 2026-10-09

This loop is the Software Factory pack's target production flow, not an integrated capability. Deltas from the 2026-10-08 packet, re-validated: separate a read-only first workflow from controlled writes, generation, delivery and product operation; show trusted submission, immutable release resolution, accepted run snapshot, scoped worker execution, model and tool mediation, authorized context, artifact validation, optional review and observable result; name one authoritative writer per state (Run Control for runs and approvals, Knowledge & Memory for artifacts; the Studio submits commands and reads projections); add denied, waiting, partial, cancel, restart and uncertain-effect branches; bind stages to FR/FUX/NFR ids rather than claiming scale. No stage of this loop runs end to end today.

The factory operates as a continuous software production loop. This page is the canonical end-to-end reference: from human intent to running SaaS and back into the knowledge graph.

The lifecycle¶

Prompt / Business Intent
→ Blueprint
→ Knowledge-Enriched Agent Workflow
→ Validated Artifacts
→ GitOps Delivery
→ Running SaaS Runtime
→ Observability Feedback
→ Knowledge Graph Improvement
→ Better Future Generation

End-to-end flow diagram¶

flowchart LR
    Intent["Human Intent / Product Goal"]
    Blueprint["Blueprints"]
    Knowledge["Knowledge Fabric"]
    Agents["Agent Mesh"]
    Artifacts["Validated Artifacts"]
    GitOps["GitOps Delivery"]
    Runtime["Generated SaaS Runtime"]
    Observability["Observability Feedback"]
    Improvement["Knowledge Improvement"]

    Intent --> Blueprint
    Blueprint --> Knowledge
    Knowledge --> Agents
    Agents --> Artifacts
    Artifacts --> GitOps
    GitOps --> Runtime
    Runtime --> Observability
    Observability --> Improvement
    Improvement --> Knowledge

Hold "Alt" / "Option" to enable pan & zoom

End-to-end sequence¶

sequenceDiagram
    actor Human
    participant Studio as Factory Studio
    participant CP as Control Plane
    participant KN as Knowledge Platform
    participant AM as Agent Mesh
    participant TL as Template & Library
    participant GOV as Governance
    participant DO as DevOps / GitOps
    participant RT as Runtime Cloud
    participant OB as Observability & Feedback

    Human->>Studio: Submit intent / product goal
    Studio->>CP: Create / update Blueprint
    CP->>KN: Retrieve patterns, prior systems, decisions
    CP->>AM: Assign agent workflow (AgentTask)
    AM->>KN: BuildContextPackage
    KN-->>AM: ContextPackage (governed, ranked)
    AM->>TL: Execute templates, evaluate compatibility
    AM->>AM: Generate & validate artifacts
    AM->>GOV: Policy / approval gates
    GOV-->>AM: Decision (approve / escalate)
    AM->>CP: Register artifacts & events
    CP->>DO: Commit, PR, pipeline, release request
    DO->>RT: Deploy generated SaaS modules (Pulumi IaC)
    RT->>OB: Emit logs, metrics, traces, incidents
    OB->>KN: Store runtime memory, feedback items
    KN-->>CP: Improvement candidates for next generation

Hold "Alt" / "Option" to enable pan & zoom

Stage-by-stage¶

Stage	Owning platform	Key artifacts/events
Intent	Factory Studio	`VisionDocument`, `ProjectCreated`
Blueprint	Control Plane	`Blueprint`, `BlueprintValidated`
Knowledge enrichment	Knowledge Platform	`ContextPackage`, `ContextPackageBuilt`
Agent workflow	Agent Mesh	`AgentTask`, `AgentTaskCompleted`
Validated artifacts	Agent Mesh + Template & Library	`Artifact`, `ArtifactCreated`, `ScaffoldOutput`
GitOps delivery	DevOps / GitOps	`PullRequest`, `PipelineRun`, `ReleasePlan`
Running runtime	Runtime Cloud	`RuntimeDeployment`, `DeploymentPromoted`
Observability feedback	Observability & Feedback	`TraceRecord`, `Incident`, `FeedbackItem`
Knowledge improvement	Knowledge Platform	`RuntimeSignal`, `KnowledgePattern`, improvement tasks

Traceability guarantees¶

Throughout the flow, actions link to traceId, agentId, skillId, moduleId, and blueprintId. Every phase emits logs, spans, events, and validation metadata. Failed flows can be retried, replayed, or stepped through via the Control Plane Workflow Replay and Knowledge replay capabilities. See the Metadata Schema for the fields that make this possible.