Zum Inhalt springen

Open Source

I work with open source every day. What comes out of my own work and proves itself goes back the same way. Below: what each tool does, what it is for, and where the code and the posts live.

NextSearch

AGPL-3.0

Self-hosted full-text search across Nextcloud folders, strictly read-only.

  • Nuxt
  • Laravel
  • Meilisearch
  • Apache Tika

Nextcloud stores and shares files but finds nothing inside them. The official full-text search can, but it wants an Elasticsearch cluster alongside that somebody has to run. NextSearch is the smaller answer to the same question: it indexes the folders you point it at — across as many instances as you like — and makes their contents searchable. PDF, Office, OpenDocument, emails as .eml, plus OCR for scans without a text layer.

Access to Nextcloud is read-only by force, and in code: the one component that talks to an instance allows four HTTP methods and throws on anything else before a socket is opened. NextSearch creates no file, changes none, deletes none.

Tags and an HTTP API arrived in 0.3.0, an MCP server in 0.4.0 — so an assistant can search and tag with the same permissions its owner has. The whole stack comes up with one docker compose up; make demo adds a throwaway Nextcloud to try it against.

proxypark

MIT

Local development domains on macOS, with trusted HTTPS and no port conflicts.

  • Traefik
  • dnsmasq
  • mkcert

Anyone working on several projects at once knows the ritual: port 3000 is taken, so 3001, then 8080, and eventually nobody remembers which application listens where. Cookies, CORS and OAuth callbacks take care of the rest.

proxypark puts a layer in front. Every project gets its own .test domain, behind it sits a Traefik pointing at the containers, and mkcert issues a certificate the machine actually trusts — a green padlock, no clicking through a warning. Resolution is handled by dnsmasq, so nothing goes into /etc/hosts.

One command per project and the domain is there. You still assign ports; they just stop being anybody's problem.

claude-code-specumentation

MIT

Claude Code skill for spec-first work: concept, epics, tickets and manuals from one source.

  • Python
  • Claude Code

In projects, specification, code and documentation drift apart the moment the first deadline bites. specumentation reverses the order: before any code exists there is a concept document under docs/concept/. From it the skill derives epics with tickets, and both stay the source that manuals, overviews and summaries are generated from as PDF.

The appeal is less in the generating than in the way back. When code changes, the documentation can be reconciled against it instead of quietly going stale. External documents — tenders, minutes, emails — can be read in and folded into the concept rather than maintained in parallel.

Meant as a convention, not a framework: a handful of fixed directories, a handful of commands, and Claude Code sticks to them.

Claude Chappe

MIT

JetBrains plugin that colours project windows by the state of their Claude Code sessions.

  • Kotlin
  • IntelliJ platform
  • Claude Code

With four IDE windows open you can no longer tell which session is working and which one is waiting for an answer. You click through, find three windows busy, and in the fourth a question that has been open for ten minutes.

Claude Chappe colours the window edge by the state of the sessions running in it: green is working, red wants something from you. That is enough to know where to go with one glance across the screen. There is also an action that stacks every project window on one screen, offset, with the most urgent one in front.

The name comes from Claude Chappe, who built the optical telegraph around 1790: signals over distance, without reading text. Roughly what the plugin does.