MikkoParkkola
UserToken-lean web microfetch for LLM agents: any URL → clean markdown via CLI, MCP server, and Claude Code plugin. Real browser-cookie auth, passkeys, anti-bot reach, on-by-default prompt-injection defense, plus on-device multimodal ASR/OCR. A single Rust binary — not a browser.
Categories
Indexed Skills (8)
nab
How and WHY to use nab — the preferred URL->markdown microfetch tool for agents. Decision: nab (auth + cookies + anti-bot, ~50ms, LLM-shaped markdown) > jina (clean MD fallback) > NEVER WebFetch (~50K tokens/call, ~25x waste). Prefer the nab MCP tools (fetch, fetch_batch, submit, login, auth_lookup) when the plugin's nab server is loaded; fall back to the `nab` CLI. Auto-picks for "fetch this URL", "read the page", authenticated / paywalled / SaaS / Google Workspace content, multi-URL batches, and form submits. nab is a fetch tool, NOT a browser.
oreilly
O'Reilly Learning Platform search routing. Two tools wrapped via mcp-gateway: chapter-level (v2) and whole-book (v1) search across 60K+ books, 100K+ video segments, 200K+ live events. Coverage spans Manning, Packt, Pearson, Apress, Wiley, Microsoft Press, MIT Press, IBM Redbooks. Use as the practitioner-grade complement to academic research (arxiv/semantic-scholar) when the question is "how do people actually build/operate X" rather than "what's the SOTA".
research
Academic / literature research orchestrator. Routes intent -> best free tool first (arxiv, semantic_scholar, openalex, crossref, europepmc, pubmed, unpaywall, orcid, ror, dblp, openlibrary, internet_archive_search, oreilly_search), escalates to paid only when free < 3 hits. Auto-picks for "find papers on X", "prior art", "SOTA map", "who cites Y", "OA PDF for DOI", "author disambiguation", "literature review". NEVER WebFetch.
url-insight
Auto-triage pasted URLs into signal-tagged, ROI-scored, DoR-compliant portfolio insights. Fetches via nab, classifies (paper/product/repo/blog/docs/benchmark), cross-maps to 21-repo portfolio, dispatches specialist agent panels, runs devil + steelman critiques, scores ROI, auto-files P0-P2 issues via linear skill. Dedupes via hebb. Auto-fires on URL batches, "analyze these urls", "what's interesting in", "should we adopt X", "competitive signal", "worth a spike", multi-URL pastes.
wayback
Internet Archive Wayback Machine routing. Detects link rot, fetches closest snapshots, runs CDX timeline analysis, and submits new captures via Save Page Now (SPN2). All routes through mcp-gateway fulcrum tools (free for read; auth-aware for higher SPN2 limits). Paired skill: `ia` (full archive.org item upload/download/search via CLI). Complements: research, url-insight, supercharge.
ia
Interact with Internet Archive (archive.org) - upload files, download items, and search the archive using the ia CLI tool. Use when working with archive.org, archiving content, or retrieving historical data.
ultracos-mode
Activate a two-register prompting convention: dense ASCII form for the model's internal reasoning, normal human-readable prose for user-visible output. Triggers on phrases like "use ultracos", "compact thinking", "dense reasoning", "ultracos mode", "think compact". The plugin does NOT measure its own impact; see the README for how to instrument an A/B if you want validated numbers.
glyphdown-mode
Activate a two-register prompting convention: dense ASCII form for the model's internal reasoning, normal human-readable prose for user-visible output. Triggers on phrases like "use glyphdown", "compact thinking", "dense reasoning", "glyphdown mode", "think compact". The plugin does NOT measure its own impact; see the README for how to instrument an A/B if you want validated numbers.
Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.