Luma Content Research

../luma-shared/SKILL.md

When To Use

• The user wants topic ideas, competitor references, or viral angles.
• The user provides a role/persona and asks what content to make.
• The user asks for an Excel-style research table.
• A workflow needs

  step0_content_research.json

  ,

  step0_content_research.csv

  , or

  step0_keywords.json

  .

Standard Flow

1. Run backend content research:
   bash

   luma-cli research run --role "<role_or_persona_description>" --mode precise --date-range 7d --output step0_content_research.json

2. Export an Excel-friendly table:
   bash

   luma-cli research export --input step0_content_research.json --output step0_content_research.csv

3. Extract keywords and topic rows:
   bash

   luma-cli research keywords --input step0_content_research.json --output step0_keywords.json --csv step0_keywords.csv

Persona Reuse

luma-cli research persona save ai_founder --role "AI tool founder making short videos for operators"
luma-cli research run --persona ai_founder --output step0_content_research.json

Agent Rules

• Keep the raw research JSON even when exporting CSV.
• Use CSV for human inspection; use JSON for agent decisions.
• Pick at most 3 high-potential references or one topic cluster explicitly before moving to rewrite.
• Research results are discovery data, not source scripts. For viral remix workflows, only download and ASR references that are likely to contain reusable spoken copy; skip non-口播 or low-signal videos. See

  ../luma-workflow-viral-remix/SKILL.md

  .
• Do not fabricate metrics that are not present in the research result.
• If research returns too few results, retry with

  --mode expanded

  or a broader role description.