b1rdmania

Dig — Data skill for Claude Code

Data community

House and techno, 1988–2008. A catalog you can search, chat to, or plug into Claude.

How to install Dig

This entry records only its repository, not the path inside it, so there is no exact command to give. Open b1rdmania/dig and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Dig does

House and techno, 1988–2008. A catalog you can search, chat to, or plug into Claude. Built on Discogs open data.

Alternatives in Data

LLM Redteam — LLM red-team corpus runner — fires categorized prompt-injection / jailbreak / system-prompt-leak / data-exfil 4.5k ★
Portaljs Add Dcat — Make a PortalJS portal harvestable by national/EU/US open-data portals — emit standards-compliant DCAT catalog 2.3k ★
Web Agent — 🔥 Open-source web data agent optimized for structured web research 1.2k ★

README

dig

House and techno, 1988 to 2008. A catalog you can search, chat to, or plug into Claude.

**[app.dig.baby](https://app.dig.baby)** · [how we built it](https://app.dig.baby/progress) · [FAQ](https://app.dig.baby/faq) · [try the pilot](https://app.dig.baby/pilot)

The story

In March I rebuilt Discogs: the full catalog, a fast API, videos wired into every page, and a trial MCP server. It worked. It also cost about $2,000 in database bills, because the full catalog is 300 GB of Postgres that never sleeps. So I parked it, kept the ideas, and rebuilt the whole thing around one trick: **the database is a build artifact.**

Discogs publishes its entire catalog every month as CC0 XML, around 200 million lines. Each cycle, a local machine ingests the full dump, streams it through SAX parsers so memory stays flat, and cuts it down with scope manifests: style allowlists, era bounds, a tier-one label list. What ships to production is a ~10 GB database of house and techno with everything cross-linked. When the scope changes, you rebuild the artifact, not the product.

Three ways in

Search: app.dig.baby. Works like Discogs: artists, labels, records. The difference: every record plays, and each label carries a curated core run of its essential records.
AI chat: /llm-beta, key-gated beta. An agentic loop over the catalog: every record it names comes from a tool call in that turn, links to its page, and renders its video. No tool result, no claim. Sessions bag up into one YouTube playlist plus a Discogs marketplace link per record.
MCP: add https://dig-mcp.fly.dev/mcp as a custom connector in Claude. Claude can then search the catalog, walk the credit and remix graph, pull label essentials, and build session playlists. No code, no API key.