VidCraft Document Analysis
Document analysis in VidCraft: parse PDF, DOCX and Markdown, derive key points, complexity, scene structure and topic suggestions. How the doc-analyzer skill works and where its heuristics end.
VidCraft Document Analysis
Purpose
With /vidcraft:doc-analyzer <file-path> [video-type] you turn existing documentation into a production plan: structure, key points, a suitable video type and a rough scene list. The analysis runs locally in the MCP server.
The file must be located under your
content_root. All analysis tools reject paths outside that directory. Copy the source to{content_root}/projects/<project>/research/if necessary.
Supported formats
| Format | Parser | Detected |
|---|---|---|
Markdown (.md) |
built in | Headings, code blocks, lists, frontmatter |
PDF (.pdf) |
pdfplumber |
Text, page structure, metadata |
Word (.docx) |
python-docx |
Headings, styles, lists, embedded images |
Images are only detected by the Word parser. For Markdown and PDF the analysis therefore never reports the "images" trait.
Extensions other than .doc are rejected with an error message. .doc is accepted and passed to the python-docx parser, but a genuine legacy .doc file raises a python-docx error there that is not caught. Convert such files to .docx first.
The five analysis tools
The skill runs these tools in order. Tools 3 to 5 run in parallel once 1 and 2 are done.
| Tool | Result |
|---|---|
analyze_document |
Title, format, word count, sections with heading, level, word count and traits (code, list, images; images for DOCX only) |
analyze_complexity |
Complexity score from 0 to 100, recommended type with reason, recommended episode count |
extract_key_points |
Up to max_points (default 10) key points with relevance and a hint on the scene kind |
suggest_video_structure |
Scene list with visual type, estimated words and duration, calculated at 140 WPM |
suggest_video_topics |
Up to max_topics (default 5) possible videos with title, type, duration and source sections |
How the heuristics work
Type recommendation (analyze_complexity), in this order:
| Condition | Recommended type |
|---|---|
| more than half of the sections contain code | tutorial |
| more than half of the sections contain lists | installation-guide |
| under 500 words | explainer |
| over 2000 words | training |
| otherwise | product-demo |
Episode count: under 2000 words one episode, up to 5000 words at least two (one third of the sections), above that at least three (half of the sections).
Scenes (suggest_video_structure): There is always an introduction at the start and a summary at the end. Sections under 20 words are skipped. Sections with code become screencast, with images split (DOCX only), with lists slides, all others avatar. The video type you pass is echoed in the result but does not change how scenes are split.
Key points: Every section with a heading yields one point. Code and lists are marked as code_demo and steps respectively, sections with images as visual (DOCX only).
These are simple counting rules. They give you a starting point, not a finished plan.
Result of the skill
The skill summarizes everything in one report: source, complexity, key points, scene structure with timings, required assets (screenshots, recordings) and next steps. On request it creates a project with create_project_structure or stores topics via create_video_idea in IDEAS.md.
For documents over 5000 words the skill recommends planning several episodes.