Skill Claude CodeCodex
Convert academic PDF papers into polished, narrative-driven PowerPoint presentations using 13 specialized AI agents orchestrated by Claude Code. Based on Rhetorical Structure Theory (RST) discourse parsing with a multi-agent critique-revise-judge refinement loop. Every slide is visually rich -- text-only slides get…
Agent
ArcDeck decomposes paper-to-slides generation into 13 specialized agents across 3 phases. Each agent is described in its own .md file that Claude Code reads and executes. The pipeline produces visually rich slides with no text-only pages -- every slide gets either extracted figures or programmatic visual elements…
Agent
Converts a PDF academic paper into clean Markdown text suitable for downstream LLM processing. This is the entry point of the entire pipeline.
Agent
Extracts visual assets (figures, tables) and bibliographic references from the PDF paper. Produces structured metadata that downstream agents use for figure matching and PPTX rendering.
Agent
Generates a "Global Commitment" — a structured contract document that defines the thesis, key takeaways, narrative spine, and slide budget for the presentation. This document is the authoritative guideline that ALL downstream agents must follow.
Agent
Constructs a hierarchical RST (Rhetorical Structure Theory) discourse tree for each paper section. Each tree captures how paragraphs relate to each other (elaboration, contrast, evidence, etc.), enabling discourse-guided slide grouping downstream.
Agent
Groups paragraphs into slides on a per-section basis using the RST discourse tree as structural guidance. Each section produces its own slide list, which are then merged into a unified presentation outline.
Agent
Reviews the merged slide plan against the Global Commitment and presentation quality rubrics. Produces actionable critique identifying issues with alignment, flow, balance, and coherence. Does NOT modify the plan — that is Agent 7's job.
Agent
Revises the merged slide plan based on feedback from the Narrative Critic (A6) and, in subsequent rounds, from the Narrative Judge (A8). The reviser may reorder slides, rename titles, move paragraphs between slides, split or merge slides, and adjust section assignments — all while preserving complete paragraph…
Agent
Evaluates whether the revised slide plan is "good enough" to proceed to slide generation, or should re-enter the critique-revise loop. The judge is a strict evaluator that primarily checks alignment with the Global Commitment.
Agent
Filters the extracted images and tables from the PDF, keeping only those that are relevant and high-quality for the presentation. Removes irrelevant, decorative, or low-quality visuals. Keeps at most 5 images and 5 tables. Also identifies slides that will need programmatic visual elements (diagrams, infographics)…
Agent
Holistic slide plan creation — simultaneously matches figures to slides, selects templates, writes content, and designs visual elements. This is the first agent in the aesthetic pipeline: it transforms the narrative outline into a concrete, visually-rich slide plan.
Agent
Reviews slideplan.json for aesthetic quality using CARP+ design principles. Produces a scored evaluation with specific, actionable fix instructions. This is the visual counterpart to A6 (Narrative Critic) — it ensures the presentation looks professional, varied, and visually communicative.
Agent
Applies design critique fixes, enriches visual elements with paper-specific content, polishes formatting, and produces the final refined slide plan. This is the last LLM-driven agent before PPTX rendering.
Agent
Assembles the final PowerPoint presentation from the refined slide plan using PptxGenJS (Node.js). A single script (generatepptx.js) handles all rendering: text with LaTeX formatting, images with aspect-ratio preservation, and programmatic visual element diagrams -- no template PPTX needed.