Workspace & delivery

A skill is a versioned instruction module that Claude Code loads at runtime — coding standards, review protocols, accessibility checklists written down once and applied the same way on every project. Zivtech maintains a library of 80+ skills; the installed working set as of June 2026 is 101 modules.

The library breaks down to 36 reviewers (critics), 33 planners, 13 executors, 4 writers and builders, and 15 other specialists. The ratio is the point: reviewers outnumber executors almost three to one, because review is the governing layer of AI-assisted delivery, not an afterthought.

The skill library itself is private today. The platform core that loads and governs skills is public, so the mechanism is inspectable even where the library is not.

Evidence-based discovery replaces assumption-driven scoping by testing the actual system before writing recommendations: every editorial form exercised, the support history read, the codebase audited — so the proposal cites observed evidence instead of pattern-matching from past projects.

On a recent research-institution engagement, that meant browser-testing all 17 editorial forms on the live site, reading six years of support history, and auditing the existing Drupal codebase before a single recommendation was written. The resulting proposal ran 12,000 words, and every recommendation traced back to something observed: a form that broke, a ticket pattern, a dependency blocking the upgrade path.

AI agents make this economically possible — they do the comprehensive, tedious part while senior people spend their hours on judgment. Joyus Recon packages the same approach: architecture discovery from a single public URL up through full codebase access.

Yes. The working pattern: Claude Code runs on every pull request via GitHub Actions, the diff is filtered to custom code only, findings are scored by severity, and results post as PR comments before any human review starts.

What makes this more than a linter is the skill layer. Skills encode Drupal coding standards, security checks, and accessibility requirements as versioned instruction modules, so the reviewer applies the same architectural judgment on every change — the things generic static analysis misses entirely.

This is not a thought experiment: the pattern runs today in continuous integration on a long-running Drupal 10 + React digital library that Zivtech maintains, with senior engineers making the final call on every finding.

Joyus Docket turns CSV and Jira backlogs into local, agent-ready work queues: triage, planning, critic review, and one-issue commits, with imported data kept as immutable truth and workflow state layered on top.

Backlogs become agent-ready work queues.

Proprietary.

Joyus Recon performs site architecture discovery and competitive intelligence: architecture reference documents, security reviews, and pre-drafted work packages, from a single public URL through full codebase access.

Understand a site — or a competitor's — before the first meeting.

Proprietary.

Joyus Context is a shared memory substrate: one provider-agnostic way for Joyus tools to store, retrieve, explain, and supersede durable knowledge across sessions and tools.

One memory layer behind every Joyus tool.

Proprietary.

Joyus Desktop is the native companion app: policy enforcement, session handoff, skill and MCP distribution, managed git sessions, and local-versus-remote environment management.

Run Joyus on your machine, under your policies.

Open source, Apache-2.0. Clone it today.