Skip to main content
The CyberOS Claude plugin is a convenience layer that packages the full task workflow as host commands. Once installed, it gives you slash commands — the five CLI-parity commands (/install, /uninstall, /version, /status, /help) plus the four task-workflow commands (/ship-tasks, /plan, /create-tasks, /improve) — directly inside Claude Code, Claude Desktop, or Cowork, without needing to copy prompts or reference file paths manually. The plugin is not required to use CyberOS; it is a shortcut on top of a doc-driven system that works with any agent.

​
Plugin commands

​
CLI-parity commands

These slash commands mirror npx cyberos <command> one-to-one.

​
Task-workflow commands

These slash commands drive the CUO task workflow. They are plugin-only — no npx cyberos equivalent.
There is no /update or /changelog command. Update is /install (re-vendor) after /version confirms a newer payload is available. The changelog lives on the Changelog page.
The CLI-only npx cyberos create, npx cyberos gates, and npx cyberos mcp commands have no plugin equivalent — see the Installation guide.

​
Install the plugin

1

Add the marketplace

Run the following in your Claude Code session, pointing at the built payload directory:
2

Install the plugin from the marketplace

Claude Code will confirm the plugin is active. All nine slash commands are now available in any repo session.

​
Human-in-the-loop is required

The plugin does not change the fundamental workflow contract. Every task still requires two recorded human verdicts before it reaches done:
  1. Gate 1 — Review acceptance (reviewing → ready_to_test): the agent presents what it built and its review findings, then halts. You read the task clauses against the change and say “approved” to continue, or route back with what is wrong.
  2. Gate 2 — Final acceptance (testing → done): the agent presents test evidence and gate output, then halts. You accept with “done” or route back.
The agent never sets ready_to_test or done on its own. An agent that claims to have done so is behaving incorrectly — report it.

​
The ship-tasks skill in depth

/ship-tasks is more than a shortcut — it is a self-contained skill that bundles its own copy of the workflow doctrine:
  • ship-tasks.md — the full workflow rules
  • EXECUTION-DISCIPLINE.md — how agents must behave between the gates
  • STATUS-REFERENCE.md — the 10 task states and who may set each
Because the doctrine is bundled inside the skill, /ship-tasks works even if .cyberos/ is not present in the current repo. The skill always carries the rules it needs.

​
Agent-independence

The Claude plugin is convenience, not a requirement. Installing CyberOS writes the canonical AGENTS.md spine, .cyberos/AGENT-ENTRY.md, and per-agent pointer files for every supported agent. Point any file-and-shell agent — Codex, Cursor, Gemini, Grok, or others — at AGENTS.md or .cyberos/AGENT-ENTRY.md, and it drives the same workflow with the same two mandatory human gates. The plugin adds slash-command access on top; the underlying system is entirely agent-independent.