Project Log: n8n Ghost Writer Automation with Local LLM (Ollama)

Summary

This entry documents the initial build and iterative refinement of the Ghost Writer automation pipeline for FlowformLab — a local, privacy-preserving content generation workflow powered by n8n and Ollama.
The objective is to accelerate first-draft creation while preserving personal narrative, structure, and editorial control.

The pipeline now accepts a topic, generates structured headings, TL;DR and diagrams, writes the article body, and merges outputs into FlowformLab-ready Markdown.

Work Timeline

Date Milestone Key Activities Output
Sat, 1 Nov 2025 Kickoff Defined goals & writing constraints. Set up n8n / Ollama baseline. Project charter + architecture outline
Sun, 2 Nov 2025 Headings Engine Built 3-heading generator sub-workflow, JSON parser + fallbacks Stable headings module
Mon, 3 Nov 2025 TL;DR + Diagram Engine Added TL;DR + Mermaid diagram subflow w/ resilience to parsing errors Summary + diagram generator
Tue, 4 Nov 2025 Body Writer + Merge Built body writer, added Markdown cleanup + JSON recovery, final merge logic ✅ End-to-end pipeline working

Total time invested: ~4 days (iterative)
Milestone achieved: Fully automated topic → structured draft flow

Deliverables Completed

  • ✅ Topic input node\
  • ✅ 3-heading generator + JSON normalization\
  • ✅ Auto-pick + manual override capability\
  • ✅ TL;DR + Mermaid diagram generator\
  • ✅ Body writer workflow\
  • ✅ Markdown assembler with safety checks\
  • ✅ Output: FlowformLab-ready draft (article_md)

Architecture Overview

[ Input ] 
   ↓
[ Headings Subflow ]
   ↓ pick
[ TL;DR + Diagram Subflow ]
   ↓
[ Body Writer Subflow ]
   ↓
[ Merge & Clean ]
   ↓
[ Draft Markdown Output ]

Risks & Mitigation

Risk Description Likelihood Impact Mitigation
Model drift LLM breaks JSON or returns unexpected formats Medium Medium Strong prompt guardrails, fallback defaults, JSON repair + post-parsers
Pipeline fragility Parsing or mapping breaks with unusual model output Medium High Unified normalizer functions + stricter schema validation
Style drift Output tone doesn’t match FlowformLab voice Medium Medium Style constraints + manual edit pass
Over-automation Losing personal narrative/voice Low High Human edit always required — tool accelerates, doesn’t replace thinking
Tooling changes Ollama / n8n upgrades changing behaviors Low Medium Version pinning + regression test prompts

Next Steps (Backlog)

Priority Task Notes
High Connect draft output to Decap CMS Commit drafts to /_drafts
High Add “style selector” input Reflective / Technical / Audit-PMO
Medium Create YAML front-matter template system Consistent FlowformLab layout
Medium Mermaid diagram refinement layer Label style consistency + readability
Low Long-form structuring mode Sections, ToC, research support

Reflection

💡 This pipeline now acts as a personal strategic writing co-pilot — not a content spawner.

I remain the author — this tool accelerates thinking and structure, but FlowformLab stays human-led.
The system proved stable, resilient to malformed JSON, and extensible — great foundations for future automation layers.

Updated: