Generate research-backed HTML slide decks from a one-line user brief
Run a six-agent court flow (Neige, Jinyiwei, Dongchang, Hanlin, Gongbu, Zhizaoju) on MiniMax or Codex
Cross-verify slide content against at least two sources before approving the deck
Reuse the AGENTS.md prompts across five different multi-agent runtimes
No code to run; you paste each AGENTS.md into a multi-agent platform and enable the suggested tools.
DaMing PPT Agent Team is a Chinese-language multi-agent setup for producing slide decks (PPT files, exported as HTML) at a high quality level. The README frames the whole thing as a Ming-dynasty imperial court: the user is the Emperor who issues the brief, and five specialised AI agents plus one coordinator do the work, with the Emperor only stepping in at two approval points. The six roles are described in a table. Neige is the coordinator, who breaks the brief into sub-tasks, hands them out, watches progress, collects the finished work, and presents it back to the user. Jinyiwei is the research agent who runs a deep web search of 3000 to 5000 characters with source URLs and credibility tags. Dongchang is the fact-check agent who cross-verifies claims against at least two independent sources and rejects work back to Jinyiwei if it finds problems. Hanlin is the content architect, who turns the verified report into a slide outline following a TED 3S principle (Story, Simplicity, Structure). Gongbu is the visual workshop, which generates images for slides marked as needing them. Zhizaoju is the PPT engineer, which uses a tool called guizang-ppt-skill to weave the outline and images into a final HTML deck. The document then walks through the full workflow as a flowchart: user brief lands at Neige, Neige dispatches Jinyiwei, the draft research is checked by Dongchang, the user approves the final research, Neige sends it to Hanlin for the outline, then Gongbu and Zhizaoju work in parallel until Zhizaoju delivers the HTML file, and finally the user gives a yes or no on the finished deck. A deliverables table lists what each stage produces and who it goes to next. A directory layout shows one folder per role, each containing an AGENTS.md file that holds the full system prompt for that agent. A platform table names five multi-agent runtimes where these prompts can be pasted in directly: MiniMax Agent, OpenAI Codex, WorkBuddy, OpenClaw, and Hermes. There is also a per-agent suggested tool list (web search for the research and fact-check roles, image generation tools like Image 2, Seedream 5, Banana 2, or Midjourney for the visual role, file and HTML rendering for the PPT engineer). The quick start asks the user to clone the repository, create the six agents in their chosen platform with each AGENTS.md as the system prompt, enable the suggested tools, then message the Neige agent with a topic and any extra requirements. The README also includes a screenshot grid of sample output decks, an About Me section from the author, and an MIT licence note.
Generated 2026-05-22 · Model: sonnet-4-6 · Verify against the repo before relying on details.