A content team needs a quick draft polished into publication-ready prose. The writer focuses on getting ideas down fast without worrying about grammar or tone; the editor focuses exclusively on refinement without the cognitive burden of generating ideas from scratch. Splitting these responsibilities produces better output than asking one person to do both simultaneously.
This is the simplest multi-agent pattern: two agents in sequence with a final assembly step that packages the complete pipeline result.
topic, tone
|
v
tap_writer_agent (draft string, wordCount=35, model)
|
v
tap_editor_agent (editedContent, notes, changesCount=4, model)
|
v
tap_final_output (finalContent, summary with metadata)
Takes topic, tone, and systemPrompt via getInputData().getOrDefault() with defaults of "technology", "professional", and "You are a skilled writer.". Generates a four-sentence draft that interpolates both the topic and tone. Returns draft (the full text), wordCount: 35, and model: "writer-agent-v1".
Receives the writer's draft and the target tone. Prepends "EDITED: " to the draft and appends a bracketed refinement note: "[Refined for clarity, coherence, grammar, and {tone} tone]". Produces four specific notes: improved sentence structure, corrected grammar, enhanced vocabulary, and added transitional phrases. Returns editedContent, notes, changesCount: 4, and model: "editor-agent-v1".
Assembles the pipeline result from originalDraft, editedContent, and editorNotes. Computes originalLength and editedLength as character counts via .length(). Returns finalContent (the edited version) and a summary map containing both lengths, the editor notes, and agentsPipeline: ["writer-agent-v1", "editor-agent-v1"] for traceability.
3 tests cover writing, editing, and final output assembly.
- RUNNING.md -- how to build and run this example