Skip to content

openknowledge_cli.md

Latest commit

 

History

History
66 lines (55 loc) · 4.14 KB

openknowledge_cli.md

File metadata and controls

66 lines (55 loc) · 4.14 KB
Raw
type Reference
title openknowledge (openknowledge-sh)
description A Go CLI for creating, connecting, inspecting, and publishing OKF bundles, with a local registry, a browser viewer, multiple exporters, and an agent maintenance loop. The most tooling-complete OKF project.
resource https://github.com/openknowledge-sh/openknowledge
tags
ecosystem
featured
okf-native
competitor
cli
registry
timestamp 2026-07-01

openknowledge (openknowledge-sh)

openknowledge-sh/openknowledge (~5★, Go) is a "CLI tool for managing Open Knowledge Format bundles" — create, connect, inspect, and publish local LLM wikis, and keep them current with a maintenance loop. It has its own domain (openknowledge.sh) and installer. Of the OKF-native cohort, it is the most tooling-complete and the most direct competitor alongside okf-skills.

What it ships

A layered command surface around a bundle:

  • Authoring / hygiene — setup, new, validate, list, spec (scaffold a bundle, seed agent maintenance rules, keep the markdown conformant).
  • Local registry — connect, disconnect, registry: give local, published, archive, or Git-remote bundles stable names that humans, agents, and the viewer resolve. This is a genuinely new idea in the ecosystem — a naming/resolution layer across many bundles.
  • Agent entrypoint — use: prints a bundle-declared instruction file (or a bundle-relative path, falling back to the root index.md) so an agent loads the right knowledge on demand.
  • Viewer — open: a registry-backed local browser UI with search and inline validation issues.
  • Export/publish — to html / --plain / to json / to graph: static viewer, plain semantic HTML, a normalized bundle model, or link-graph JSON. Published exports include an openknowledge.json manifest + a .tar.gz archive so connect <url> can materialize a remote bundle into a local cache.

Ideas worth stealing

  • Agent-run setup via a printed prompt. openknowledge setup prints an agent prompt rather than scaffolding directly; you paste it into Claude/Codex/Cursor (or claude "$(openknowledge setup)"). The agent inspects the workspace + memories, asks only missing questions, then builds a use-case-tailored bundle. A clever division of labor: deterministic CLI for structure, agent for judgment — directly relevant to how our ingest/init skills should behave.
  • Registry / named bundles across sources (local, published, archive, Git) — a multi-bundle-workspace idea like the OKF-native agent's multi-kb model, but generalized to remote resolution.
  • Pins a copy of the spec into every new bundle, and keeps its own docs (Wiki/) as an OKF bundle with decisions/, changelog/, workflows/ — more dogfooding with an explicit decisions/workflows split worth imitating.
  • "Focused lower-reasoning subagents on bounded wiki-maintenance tasks" — its setup guidance tells the host agent to delegate narrow maintenance to cheaper subagents. A concrete cost pattern for the token-cost critique.

Relevance to us

openknowledge and okf-skills bracket our space: okf-skills is Claude-Code-native skills; openknowledge is a language-agnostic CLI + registry + viewer driven by agent prompts. Both lean toward authoring/validating/publishing. Our distinct value remains the incremental, trust-modeled ingest loop from raw sources and the schema-layer portability seam. Its use-prints-an-entrypoint and setup-prints-a-prompt patterns are strong, concrete ideas to borrow for our skills' ergonomics. It is also a second independent consumer/validator (like kiso) to interop-test our output against.

Citations

  1. openknowledge — https://github.com/openknowledge-sh/openknowledge