html

# HTML Use this command-like skill for `/html` / `Lumina HTML`. `/html` turns an existing Autody run folder into a Lumina HTML page. It is a rendering workflow, not a collection workflow. Regenerate the report analysis from the latest data every run. Do not use Chrome, Doubao, creator center, or public Douyin pages from this command. If required data is missing, report the exact missing files/fields and, when the user's current request is actually collection, update, backfill, or audit, route yourself to the matching Autody workflow in the same turn instead of telling the user to run it. If the request is strictly `/html`, render only when the local data is sufficient. ## Required Context Before rendering, read: 1. `../douyin-analysis/references/report-design.md` 2. `../douyin-analysis/references/report-agent.md` 3. `../douyin-analysis/references/lumina-html-workflow.md` Use Lumina only. Do not offer a three-template picker and do not mention unused template families. Use the Guizang reuse pattern: keep visual elements stable, but rebuild the view model and conclusions from current data. Old Lumina or Guizang HTML files are visual references only; never copy their old conclusions into a new report. ## Inputs Use the target run folder, normally: ```text outputs/douyin_analysis_YYYY-MM-DD/ ``` Prefer: ```text douyin_deep_works_final.json douyin_incremental_analysis_YYYY-MM-DD.json content_gap_audit.json ``` ## Output Write a fresh payload and Lumina HTML file into the run folder: ```text report_lumina_payload.json report_lumina.html ``` Use: ```bash node ~/.codex/skills/douyin-analysis/scripts/render_lumina_report.cjs \ --works outputs/douyin_analysis_YYYY-MM-DD/douyin_deep_works_final.json \ --analysis outputs/douyin_analysis_YYYY-MM-DD/douyin_incremental_analysis_YYYY-MM-DD.json \ --audit outputs/douyin_analysis_YYYY-MM-DD/content_gap_audit.json \ --out outputs/douyin_analysis_YYYY-MM-DD ``` The payload must include `generatedAt`, `sourceFiles`, `summary`, and `items`. Report the HTML path and any data quality caveats. If the payload includes new videos, carry through `blindScoreStatus`, the blind prediction, observed result, delta, calibration, and `blind_score_blocked` state from the report analysis. If `blind_score_blocked`, show that status instead of inventing a blind prediction. Do not create retrospective blind scores during rendering. ## Boundaries - Do not collect or backfill browser data. - Do not generate transcripts. - Do not inspect cookies, localStorage, passwords, session stores, or browser profile files. - Do not offer a three-template picker. - Do not invent strategic conclusions when the input data does not support them. - Do not reuse stale report conclusions. Reanalyze against the current source files before rendering.