Turn research findings into a tested writing instruction file
At a glance
A writer used NotebookLM to study a paper, translated its findings into editing rules, and revised the resulting AI skill through writing tests. This workflow is listed under Agent Instructions, Writing, Research, Evaluation.
- Agents / tools
- NotebookLM, AI writing assistant, Markdown
SourceBuilding a tool for these workflows? See weekly rates.
What they did
1) Have NotebookLM, a research paper about writing, a text editor, and an AI writing assistant that accepts supplied instructions. Upload the paper to a notebook. 2) Ask which principles can become practical editing rules, which patterns cause repetition, and what examples the paper supplies. Use follow-up questions to separate actionable guidance from academic background. 3) Write `skill.md` with sections for the assistant's role, principles, editing rules, audience modes, unwanted patterns, editing sequence, and a final checklist. Translate findings into direct instructions and add before-and-after examples. 4) Give the assistant those instructions with sample blog, academic, technical, and marketing text. Inspect whether tone becomes too casual, terminology disappears, or paragraphs become formulaic. Revise the rules and retest; retain an explicit requirement to preserve claims, evidence, citations, and qualifications. Save the revised instruction file for reuse.
How it went
The author says testing exposed tone and terminology problems and that the article itself was written using the resulting skill.
