AI workflow · Evidence first
Turn an industry question into a report people can actually read.
Research to HTML is a repository-scoped Codex skill for evidence-backed market and industry research. It keeps the research trail reviewable, then turns the approved findings into separate English and Simplified-Chinese static pages with the same conclusions, caveats, and sources.
- Primary output
- Bilingual static HTML
- Evidence source
- Reviewable Markdown
- Optional release
- GitHub Pages
One decision artifact, three quality layers.
The workflow treats research, translation, and interface design as connected checks—not as three unrelated generation steps.
Traceable evidence
Current claims point to direct sources. Observations, inferences, unknowns, dates, geography, and confidence stay distinguishable.
Equivalent bilingual pages
English and Chinese pages share the same figures, causal claims, limitations, interactions, and source set.
Low reading pressure
The conclusion appears first. Clear hierarchy, reading controls, accessible interactions, print support, and responsive layouts reduce effort without hiding evidence.
Use it for market structure, product and platform comparisons, business-model questions, creator economics, or technical ecosystems where the answer must remain inspectable after the AI session ends.
From question to public report.
The skill follows a fixed evidence-to-publication sequence while adapting the analysis to the domain.
- Define the decision. Set the audience, scope, geography, research cutoff, operational terms, and exclusions.
- Build the evidence record. Prefer current primary sources, record caveats, and separate what is observed from what is inferred.
- Write before styling. Approve the Markdown findings before turning them into English and Chinese reader interfaces.
- Validate the public artifact. Check privacy, citations, locale parity, accessibility, responsive behavior, metadata, JavaScript, print, and the exact files intended for publication.
Inspect the output before adopting the workflow.
The first published report tests a real two-part decision across Chinese and international crowdfunding platforms.
Crowdfunding products and indie games: profit structure, platform fit, and game types
See how the report distinguishes money raised from money retained, tests the publicity and funding value of crowdfunding for indie games, and exposes both evidence and limits.
Clone, open, ask.
The skill lives in the repository’s standard agent-skill directory, so Codex can discover it when the repository is open.
- Clone or download the Research repository.
- Open the repository in Codex.
- Send the prompt below and replace the bracketed question.
Use $research-to-html to research [your industry decision] and create a bilingual web report.
What the workflow does not pretend to know.
Reliable output depends on evidence quality, a defined decision, and review—not on the confidence of generated prose.
Revenue is not profit
The skill does not infer profit from pledges, GMV, traffic, downloads, or funding success.
Unknown stays unknown
Missing platform data, unaudited claims, stale examples, and causal uncertainty must remain visible in the report.
Publishing is a separate gate
Only an explicit reader-facing allowlist is deployed. Skills, scripts, notebooks, templates, credentials, and local configuration remain outside the Pages artifact. Use research-to-github-pages as the optional second step; it does not need a separate product page.