From a Website Promise to a Working Harnez Catalog

Harnez.org is the public-facing home for a developer's agentic coding harness: a set of tools and repeatable workflows for working across repositories with coding agents. The site had a discovery problem. Its navigation promised visitors places to find tools, skills, and commands, while the catalog did not yet give those categories a useful, browsable inventory. The work recorded in September 2026 turned that gap into a concrete static website feature.

The repository builds the site from a YAML specification. That specification describes tools, practices, standards, documents, installation guidance, navigation, and workflow entries. A Node.js build script validates the specification and generates static HTML pages. The browser experience does not depend on a server or third-party JavaScript; the catalog filter is a first-party progressive enhancement. The build uses the system cmark program to render reference documents. The documented checks validate generated content and links, reject certain runtime features, and probe pages through a local static HTTP server.

The central catalog milestone landed on 2026-09-14. Commit 2b9e21d refreshed tool information and added skills, commands, and installation pages. The implementation expanded the schema, catalog generator, browser filter, and many generated routes together. It was followed by 4fcee93 and e18c190, which closed issues 007 and 018 after recording that the formerly empty categories were populated and static, browser, and source checks passed. Two days later, commit 43c6913 refreshed the site's Harnez capabilities and workflow catalog. These are practical steps in turning an internal workflow inventory into pages an outside visitor can navigate.

One telling design decision was to model related skills and commands as shared workflow entries. The site expands those registrations into separate searchable pages, while keeping their descriptions and prerequisites together in the source specification. This makes an important distinction visible: a workflow can be installed as a skill, invoked as a command, or registered in both ways, and the catalog should reflect those actual registrations instead of implying that every workflow has the same form. The docs also call out that the catalog is a curated public selection, not a complete dump of internal configuration.

There is evidence of disciplined checking around that expansion. Issue 018 describes the inventory research across Harnez and sibling projects, calls for verifying releases and setup commands, and warns against inventing standalone installers for workflows that do not have them. Its closure records static, browser, and source checks. Earlier work on the standalone site also tracked issues for unsafe or misleading Markdown rendering, stale static-host assumptions, inaccessible navigation, and missing trust signals. A public catalog is not only a collection of cards: it has to stay accurate as the tools and their installation paths change.

This repository demonstrates agentic development most clearly as an operating environment and a body of evidence, rather than as proof that any one feature required a team of agents. In the six weeks ending 2026-10-04, Git records 20 commits, including the catalog implementation, generated-site updates, issue lifecycle changes, and five research documents on agent skills, harnesses, evaluation services, real-world engineering evidence, and Harnez's direction. The project documents issue-driven work, canary-first checks, sprint and lean-sprint workflows, advisors, and tool feedback. Its catalog even explains delegated sprint workflows. Those facts show that the repository is structured to support agentic work; the commit history alone does not establish that all of those mechanisms were used to produce the site changes.

The interesting lesson is the combination of breadth and restraint. A small static site can describe a wider tool ecosystem, but useful agent-produced catalog pages need provenance, accurate prerequisites, and checks that reflect how visitors will encounter the result. Harnez.org's September work made more of the ecosystem discoverable while keeping the source inventory explicit and the published experience static. The remaining open issues are a reminder that a functioning catalog is a milestone, not the end of content and accessibility maintenance.

Facts