Intake

The intake layer turns source material into plain text before the readiness gate scores it. It should extract text, not decide whether the source deserves a skill.

Supported Inputs

Input	Status	Notes
Local .txt	Supported	Read as UTF-8 text
Local .md / .markdown	Supported	Preserves Markdown structure
Local .html / .htm	Supported	Removes script, style, head, nav, and footer content
Local .epub	Supported	Reads package metadata and spine XHTML chapters
Remote http / https HTML	Supported	Fetches one URL and applies the HTML cleaner
Remote http / https text	Supported	Fetches one URL and reads text content
SRT / VTT transcripts	Supported through clean-transcript	Clean first, then analyze the Markdown output
Audio files	Supported through transcribe-audio when a Whisper-compatible CLI is installed	Transcribe first, then clean and analyze

URL Usage

source-to-skill analyze https://example.com/article
source-to-skill build https://example.com/article --level auto --out out

URL intake is intentionally modest. It fetches a single URL with a normal user agent and uses the same conservative HTML cleaner as local HTML files.

EPUB Usage

source-to-skill analyze path/to/book.epub
source-to-skill build path/to/book.epub --level auto --out out

EPUB intake reads META-INF/container.xml, the OPF package file, metadata title, and spine XHTML chapters. It does not attempt DRM removal or publisher-specific layout reconstruction.

Not Yet Supported

• multi-page crawling
• JavaScript-rendered extraction
• login-gated pages
• PDF parsing

Those should be added as intake plugins that normalize sources into text before the readiness gate runs.