Skills Explorer: pricing, features, alternatives

Skills Explorer is a website and app that indexes agent SKILL.md files from GitHub and draws an interactive visual flow for each skill. The live site is skillexplorer.dev. The repo freeflow-community/skill-explorer is TypeScript, created 2026-09-19. On 2026-09-20 GitHub listed 1 star and 0 forks. 1 star is a heat signal, not traffic. The GitHub API did not return an SPDX license on that date; do not assume MIT. The site title is "Skills Explorer". The Reddit post that surfaced it described a visual flow so you can see what a skill does without reading a wall of LLM prose.

Home shows newest skills, a tag cloud, collections, and search over names and descriptions. A skill page lists repository, path, branch, collection, and tags, then the flow. If the flow is missing, Build visual flow starts a background job and the page polls until it is ready. Compare Agent Skills if you wanted a pack to install, Agent Dispatcher if you wanted a router over a local library, or Claude Code if you wanted the host rather than a catalog.

Key Features

Register from GitHub: npm run cli -- register owner/repo with optional --path, --collection, --tags, or --auto-tags. --auto-tags asks the configured model for 3-6 tags and reuses existing ones when it can. The same repo plus path updates in place.
Visual flows: src/flows/generator.ts pulls SKILL.md and text resources (up to FLOW_SOURCE_BUDGET characters) and sends them to Anthropic Messages (default model claude-opus-5) or OpenAI Responses (default gpt-5.4-mini). FLOW_GENERATOR=stub builds a placeholder outline with no API key. Instructions live in prompts/visual-flow.md.
Storage: one Cloudflare R2 bucket, or ./data/blobs when R2_BUCKET is unset. The index is SQLite (index/skills.db). Flows are flows/<slug>.html plus JSON metadata (model, build time, source files, token usage). The server never writes the index; sync push / sync pull move a snapshot. A push is refused if someone else pushed in the meantime unless you --force.
Sandbox: generated flow pages are third-party model output. They are served with a sandboxing Content-Security-Policy and shown in a sandboxed iframe. They may run their own scripts and load Google Fonts, and nothing else. Set FLOW_BUILD_TOKEN if you want to require a token before a paid build.
Local run: Node 24+. npm install, optional .env, npm start on http://localhost:8787. npm test is a node:test suite with no network.

Limitation: building a real flow calls a paid model unless you use stub mode. The catalog is whatever operators registered; it is not every skill on GitHub. Suggesting a skill from the site opens a GitHub issue, it does not write the index from the browser.

Use Cases

Browsing unfamiliar skills before you npx skills add anything.
Self-hosting an internal catalog with R2 or local blobs.
Skip if you already know the three skills you use. A directory is overhead.

Pricing

Piece	Price	Notes from first-party pages 2026-09-20
Site / repo	$0 to clone	License not stated on GitHub as of 2026-09-20. About 1 star.
Hosted skillexplorer.dev	Not priced on the homepage	No plan table on the public page.
Flow builds	Your Anthropic or OpenAI bill	Default generators call those APIs. Stub mode is free of model spend.
R2	Cloudflare's object storage rates	Optional; local `./data/blobs` if unset.

Getting Started

Open skillexplorer.dev and search, or clone the repo.
Local: Node 24+, npm install, cp .env.example .env, npm start.
npm run cli -- register anthropics/skills --auto-tags to index a repo of SKILL.md files.
Open a skill page and build a visual flow, or npm run cli -- build-flow <slug>.

First-party start: the README.

Frequently Asked Questions

Is every Claude skill in the index?

No. The index is the SQLite file operators registered and synced. Use the Add Skills issue link to suggest more.

Do flows execute the skill?

No. A flow is a generated walkthrough of the skill text, sandboxed in an iframe.

Can I run it without API keys?

Yes. Leave .env empty for local-only storage, and set FLOW_GENERATOR=stub if you only need UI outlines.

Alternatives

Agent Skills: installable skills, not a browser catalog.
Agent Dispatcher: routes a local library; does not draw flows.
Claude Code: where most of these skills actually run.

Tips

Treat auto-tags as model suggestions. Edit with npm run cli -- tag.
Do not --force a sync push unless you intend to overwrite a remote index.
Keep FLOW_BUILD_TOKEN on any host that is reachable beyond localhost.

Conclusion

Skills Explorer is a new, tiny catalog for people who want to see a skill as a flow before they install it. Use the public site to browse, or self-host if you need an internal index. Do not treat one star or a missing SPDX license as a maturity signal.