agentic

@rexeus/agentic

A multi-agent development toolkit for Claude Code and OpenCode. Eleven specialized agents, one orchestrator, zero complexity.

Why

Most Claude Code plugins are overwhelming. Dozens of commands, complex configuration, steep learning curves.

Agentic is different. Install it, type a command, and let the agents do their work. The Lead orchestrator figures out which specialists to deploy, briefs them precisely, and synthesizes their results. You stay in control — the agents stay in their lane.

Built for TypeScript projects. Should work with other languages too.

Quick Start

OpenCode (recommended)

npx @rexeus/agentic install opencode

One command. This installs the Agentic plugin, agents, commands, and skills globally. Restart OpenCode, switch to the visible lead agent, then run:

/agentic-plan      Plan a feature
/agentic-develop   Build it
/agentic-review    Review the code
/agentic-simplify  Make it simpler
/agentic-polish    Harmonize the codebase
/agentic-verify    Run the quality gate
/agentic-commit    Commit with Conventional Commits
/agentic-pr        Create a Pull Request

Claude Code

/plugin marketplace add rexeus/agentic
/plugin install agentic@rexeus

The Workflow

A typical development cycle with Agentic:

Plan → Develop → Review → Simplify → Verify → Commit → PR
                                ↑
                              Polish (iterative loop)

1. Plan. Start with /agentic-plan (OpenCode) or /agentic:plan (Claude Code). The Lead doesn't just accept your requirements — it challenges them. It asks hard questions, surfaces assumptions, and presents options before producing an implementation plan. You approve the plan before any code is written.

2. Develop. /agentic-develop (OpenCode) or /agentic:develop (Claude Code) runs the full pipeline. The Lead scouts the codebase, designs the approach, hands a precise briefing to the developer agent, and follows up with review and tests. You get working, tested code — not a plan about a plan.

3. Review. /agentic-review (OpenCode) or /agentic:review (Claude Code) deploys six specialists in parallel: three reviewers (correctness, security, maintainability) and three testers (coverage, craft, testability). Each has its own identity and its own loaded skills. All six are advisory — none of them modifies files. High-confidence findings only, lens labels preserved. Composite verdict is the worst of the six.

4. Simplify. /agentic-simplify (OpenCode) or /agentic:simplify (Claude Code) is where the craft happens. The Refiner distills working code to its essence — fewer abstractions, clearer names, less indirection. Behavior stays the same. Complexity goes down. This step is what separates code that works from code that sings.

Polish. /agentic-polish (OpenCode) or /agentic:polish (Claude Code) is the consistency loop. It discovers the patterns your project already uses, finds where peer files diverge, and unifies them. Use it after a feature is built, after a large refactor, or whenever files have drifted apart. Polish is designed for iterative runs: execute, review the changes, then /clear and run it again. Each pass finds fewer issues until the codebase converges.

5. Verify. /agentic-verify (OpenCode) or /agentic:verify (Claude Code) is the pre-ship quality gate. It runs the reviewer trio and the tester trio — six specialists in parallel — plus a full test-suite execution. One command, six perspectives, a clear verdict: PASS, FAIL, or CONDITIONAL.

6. Commit & PR. /agentic-commit + /agentic-pr (OpenCode) or /agentic:commit + /agentic:pr (Claude Code) handle commit and PR flow. commit creates Conventional Commits from your staged changes. pr crafts a Pull Request with a structured description. You stage the files — the agents handle the message.

OpenCode CLI

agentic install opencode
agentic doctor
agentic uninstall opencode

You don't have to use every step. Skip what you don't need. The commands work independently.

The Agents

Eleven cognitive modes, one orchestrator. Each answers a different question: