Give Claude Code an engineering process.
Claude Code is astonishing in week one and exhausting by week six: decisions nobody wrote down, inconsistent patterns, architecture that exists only in old chat sessions. Ultra Claude is a plugin that fixes that with a process rather than a better prompt — researched plans, enforced standards, agent-reviewed and tested code, and documentation that stays true to the codebase.
Full documentation: https://ultra-claude.dev
Two commands in any Claude Code session:
/plugin marketplace add duniecdawid/ultra-claude
/plugin install uc@ultra-claude
Every skill becomes a /uc:* slash command. Optionally keep it current automatically:
/plugin marketplace auto-update duniecdawid/ultra-claude
Once per machine. Installs prerequisites, configures your shell for agent teams, and sets up tmux if you want the visual pane layout. Safe to re-run.
/uc:setup
Once per project, run from the project root. It reads your codebase, scaffolds
documentation/, and derives architecture docs and coding standards from the patterns
already there. Works on existing codebases, not just greenfield.
/uc:migrate
Review what it generated — documentation/technology/architecture/ and
.../standards/. Every planning skill reads these, so the more accurate they are, the
better the plans.
/uc:feature-mode add user authentication
Feature mode behaves like a senior tech lead: it challenges your scope, researches the codebase in parallel, and produces a structured plan with testable tasks — before any code is written. You approve the plan, then:
/uc:plan-execution {plan-name}
Each task gets a dedicated team: an Executor that writes the code, a Reviewer that enforces your standards, and a Tester that validates against the product docs. With tmux, they appear as live panes you can watch. Sessions can die and resume — checkpoints save progress.
Run /uc:help any time to find the right command for what you are doing.
Not a framework, library, or runtime. It is 23 skills and 11 agents packaged as a Claude Code plugin, running entirely inside Claude Code on your machine. Nothing phones home, there is no account, and you pay only your normal Claude usage.
The documentation it writes is plain markdown that stays useful on its own. Uninstall the plugin and your repository still works.
The idea it is built on: code is a derived artifact. The specification is the source of truth, architecture docs exist before code is written, and when code diverges from the spec you fix the spec first. Documentation acts like zoning law — you build freely within the constraints, but the constraints control direction. Additive changes flow freely, compatible changes get lightweight review, breaking changes require updating the architecture doc first.
| Page | |
|---|---|
| https://ultra-claude.dev/getting-started | Install, setup, and first feature in 10 minutes |
| https://ultra-claude.dev/docs/discovery | Product research and market analysis |
| https://ultra-claude.dev/docs/feature-planning | Scope challenge, research, structured plans |
| https://ultra-claude.dev/docs/plan-execution | Agent teams, task pipeline, checkpoints |
| https://ultra-claude.dev/docs/debugging | Hypothesis-driven bug investigation |
| https://ultra-claude.dev/docs/verification | Doc-code drift detection and fixes |
| https://ultra-claude.dev/docs/standards | Define, enforce, and verify coding standards |
| https://ultra-claude.dev/docs/reference | Every skill and agent |
Bug reports and fixes are welcome — see CONTRIBUTING.md for the repository rules a pull request has to satisfy, and Discussions for questions, ideas, and showing what you built.
FSL-1.1-ALv2 — the Functional Source License. Use it, modify it, fork it and contribute back freely. The one thing you may not do is sell a competing product built on it. Every version converts to Apache 2.0 two years after release.