AI Content Pipeline
Bundled preset wf-ai-content-pipeline — multi-stage LLM and agent nodes with a brand/policy gate.
Preset id wf-ai-content-pipeline. Stages include research, outline, draft, editorial, fact-check, normalization, a condition gate, and publish or revision paths.
LLM and agent nodes need a configured provider (keys in integration_configs or environment) for real inference; without that, executors may record placeholder output depending on tier.
Use Run with sample inputs in the builder to enqueue with a small demo context object.
See External APIs for connector patterns.
Trigger shape (sample payload)
{ "sampleTopic": "cascades workflow automation", "sampleAudience": "platform engineers" }Node-by-node (abbreviated)
- Trigger — ingress.
- Research Gather —
internalEcho→/api/workflow-echo(deterministic “external” step). - Outline → Draft → Editorial → Fact-check — LLM/agent executors (placeholders without provider keys).
- Normalize / Brand gate — condition; TRUE → publish integration gate.
- CMS / publish integration? —
hasIntegration:cms-publish; FALSE → internal stub instead of live publish API.
Run with sample inputs
workflow_id: "wf-ai-content-pipeline" with the context object above.
Expected logs
LLM stages log task_completed with stubbed output when providers are absent. No integration_missing on guarded publish when cms-publish is absent.
Proof summary
Use successfulTaskCount / failedTaskCount / integrationFailures on the proof metadata JSON.