obscura-scraper-crawler

NoiseMeld Skills

Agent skills for extracting documents, editing prose, and structured critique. Built on the Agent Skills open standard, so they install cleanly into Claude Code, Cursor, Gemini CLI, Goose, OpenCode, Windsurf, and other compatible agents.

Installation

Skills CLI (cross-agent, recommended)

One install command, every supported agent picks it up:

npx skills add https://github.com/NoiseMeldOrg/skills --skill extract-book

Swap extract-book for extract-study, extract-transcript, extract-webpage, obscura-scraper-crawler, clear-and-concise-humanization, or accountability-panel. Add -g to install globally, or omit for project-only.

Update later with npx skills update. List installed skills with npx skills list.

Claude Code plugin marketplace

Register the marketplace once:

/plugin marketplace add NoiseMeldOrg/skills

Install the skills you want:

/plugin install extract-book@noisemeld-skills
/plugin install extract-study@noisemeld-skills
/plugin install extract-transcript@noisemeld-skills
/plugin install extract-webpage@noisemeld-skills
/plugin install obscura-scraper-crawler@noisemeld-skills
/plugin install clear-and-concise-humanization@noisemeld-skills
/plugin install accountability-panel@noisemeld-skills

Or grab a bundle:

/plugin install extraction-skills@noisemeld-skills    # all four extract skills

Plugin skills are namespaced (/noisemeld-skills:extract-book) and available in every project.

Manual install (no namespace)

Clone and symlink for shorter /extract-book names:

git clone https://github.com/NoiseMeldOrg/skills.git ~/skills

ln -s ~/skills/skills/extract-book ~/.claude/skills/
ln -s ~/skills/skills/extract-study ~/.claude/skills/
ln -s ~/skills/skills/extract-transcript ~/.claude/skills/
ln -s ~/skills/skills/extract-webpage ~/.claude/skills/
ln -s ~/skills/skills/obscura-scraper-crawler ~/.claude/skills/
ln -s ~/skills/skills/clear-and-concise-humanization ~/.claude/skills/
ln -s ~/skills/skills/accountability-panel ~/.claude/skills/

Note for accountability-panel: the skill is meant to be customized — personas.md next to SKILL.md is where you replace the four shipped defaults with real people whose judgment you trust. If you symlink, edits land in this repo. If you'd rather keep your customized personas.md private, copy the skill into ~/.claude/skills/accountability-panel/ instead of symlinking and edit there. See the privacy note at the bottom of personas.md.

Pull the repo to update. Symlinks pick up changes immediately.

Project-level install

Add a skill to one project so your team gets it through version control:

ln -s ~/skills/skills/extract-study /path/to/project/.claude/skills/

Python dependencies

The extract skills need Python packages:

python3 -m venv .venv && source .venv/bin/activate
pip install pdfplumber youtube_transcript_api trafilatura playwright
playwright install chromium

extract-book and extract-study use pdfplumber. extract-transcript uses youtube_transcript_api when fetching from YouTube URLs (pasted transcripts need nothing). extract-webpage uses trafilatura for static HTML and falls back to playwright (with a bundled Chromium) for JavaScript-rendered pages.

obscura-scraper-crawler is a separate path for sites behind Cloudflare/bot walls. It needs trafilatura, readability-lxml, markdownify, lxml, playwright, and the obscura binary on PATH. It uses Playwright purely as a CDP client to drive obscura serve, so you do NOT need playwright install -- obscura ships its own browser engine.

---

Skills

extract-book

Converts a PDF book into Markdown with chapters, metadata, and cleaned text.

Give Claude a PDF path, or run /extract-book path/to/book.pdf. Claude starts with a dry run to preview detected chapters, reviews the results with you, then extracts and post-processes: fixes the auto-detected title, fills in any missing metadata, and handles image placeholders.

The bundled script detects chapters four ways: CHAPTER N text markers, bare number pages, ALL-CAPS section headers, and table-of-contents matching. It extracts author, publisher, copyright, and ISBN from the first pages and strips page numbers and watermarks.

Flags: --dry-run to preview, --render-images to capture image-heavy pages as PNGs, -o path.md to set the output path.

---

extract-study

Converts a research paper PDF into Markdown with IMRaD sections, metadata, and references.

Give Claude a PDF (with an optional PubMed or DOI link for context), or run /extract-study path/to/paper.pdf. Claude runs a dry run, extracts, then verifies the title, authors, and DOI, writes 3-6 key findings bullets, and spot-checks tables for column-merge artifacts.