From ariadne-core
Interactive walkthrough of Ariadne Core. Triggers: what is ariadne core, tell me about ariadne, explain ariadne, ariadne overview.
How this skill is triggered — by the user, by Claude, or both
Slash command
/ariadne-core:ariadne-core-walkthroughThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
DO NOT use this skill when the user wants to deploy (use ariadne-core-deploy),
README.mdassets/images/add_marketplace.pngassets/images/customize_claude_desktop.pngassets/images/sync_marketplace.pngassets/images/video_thumbnail.pngevals/evals.jsonimage_manifest.yamlkiss_audit.txtproject_knowledge_graph.yamlreferences/ConceptViz_prompts.mdreferences/audience_agent_builder.mdreferences/audience_developer.mdreferences/audience_evaluator.mdsaving_tokens_transcript.txtstarter_deck.mdstupid_button_prompt.txttoken_translator.txtvalidate_manifest.pyDO NOT use this skill when the user wants to deploy (use ariadne-core-deploy), modify the codebase (use ariadne-core-build), process or search documents (use ariadne-document-intelligence), or explicitly names a different skill (use ariadne-core-router).
This skill runs in Claude Code Desktop. It does not work in Cowork.
The walkthrough uses the preview panel to display rich HTML content with
images, while questions go in the chat via AskUserQuestion. A local Python
HTTP server serves static files from walkthrough_html/.
preview_start with name "walkthrough" (defined in .claude/launch.json).starter_deck.md from this skill directory silently (do not show it).Your first user-facing message is Beat 1. No warm-up, no overview, no introduction. Beat 1 IS the opening.
You do NOT:
https://github.com/denson/ariadne-coreEvery beat follows this exact sequence:
preview_eval → window.location.href = 'http://localhost:8901/beatN.html'AskUserQuestion with 2-4 options.Never deliver two beats without a user response between them.
Beat 1 — The Hook: http://localhost:8901/beat1.html
Nate Jones's argument — frontier rates for junk bytes, 20x reduction, we built
the pipeline. Mention that Ariadne Core is a deployed service, not a local
library — trying it requires a Railway account (~$5/mo, free tier available) or
any Docker host, and takes about 5 minutes. This sets expectations before the
user tries to use tools that don't exist yet. Ask if they've seen the video.
Beat 2 — The Problem: http://localhost:8901/beat2.html
Two mechanisms of waste. Mechanism 1: raw PDF bloat. Mechanism 2: the LLM
extraction loop. Better, not just cheaper. Ask if this matches their workflows.
Beat 3 — Who Are You?: http://localhost:8901/beat3.html
Subscription = runway, agentic = cost reduction. Same mechanism, different
experience. Ask which sounds like their situation.
Based on the user's answer, branch into dynamic beats. For each:
walkthrough_html/dynamic_N.html using style.css and the
template from starter_deck.md.assets/images/ (in this skill directory) into
walkthrough_html/.AskUserQuestion. Stop and wait.Path A — Self-host on Railway: Check if they have a Railway account (free tier
to try, ~$5/mo hobby plan). Deploy walkthrough → hand off to ariadne-core-install.
Path B — Don't want to manage infra: Managed version coming → interest capture.
Path C — Local MarkItDown only: Extraction-only setup, pip install markitdown.
No deployment needed — contrast with the full pipeline which requires hosting.
docs/TOKEN_SAVINGS_FRAMING.md verbatim. Never invent figures.AskUserQuestion), content in the preview panel.starter_deck.md — 3-beat structure + exit paths + dynamic templateimage_manifest.yaml — image descriptions, topics, audience tagsproject_knowledge_graph.yaml — concept entries for dynamic beatsassets/images/ — full 26-image setreferences/ — audience-specific angle documentssaving_tokens_transcript.txt — Nate Jones video transcriptstupid_button_prompt.txt — diagnostic prompt for session wastetoken_translator.txt — per-session token math promptConversational but substantive. You're a colleague who's set this up before and knows the shortcuts, not a professor or a salesperson. Respect the user's intelligence. Be practical: move them from "what is this?" to "I have it running" as efficiently as possible.
Creates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.
npx claudepluginhub denson/ariadne-core