@jetrabbits/agentic

agentic — one AI engineering team for every coding agent

Adaptive SDLC setup for Claude Code, Codex and OpenCode. Native agents by default; workflows, helper agents and model routing only when the task needs them.

Website · Releases · npm · Usage guide · Reference

The problem it solves

Teams now use several coding agents at once. Each one gets its own prompts, its own habits and its own idea of "done". The result is uneven: one agent writes tests, another skips them; one asks before touching infrastructure, another does not.

agentic gives every agent the same playbook:

• 🧭 One process. A product owner, team lead, developer and QA agent hand work to each other with clear acceptance criteria — in Codex, Claude Code and OpenCode alike. The default lean profile keeps a short core and a quality gate; --profile team installs the full role team.
• 🛡️ Shared guardrails. 85 rules and 73 workflows for backend, frontend, DevOps, data, ML, security and more, loaded only when the task needs them.
• 🔁 Install once, keep in sync. agentic install writes the native files each agent expects; agentic upgrade keeps them current without overwriting your own edits.
• 🔔 Know when work is done. Optional Telegram messages when an agent finishes or waits for you.

Runs on the agents you already use

	Agent	Status	Evidence
✅	Codex	Real runs	0.163.0 · 13 pilot runs + live proof of JEV model routing, 2026-10-20
✅	Claude Code	Real runs	2.1.304 · 13 pilot runs, 2026-10-20
✅	OpenCode	Real runs	2.0.29 · 13 pilot runs, 2026-10-20
☑️	Antigravity	Configs ship	Agents, rules and workflows are installed; no recorded run yet
☑️	Gemini	Configs ship	Agents and workflows are installed; no recorded run yet
☑️	Cursor · Kilocode	Configs ship	Rules and MCP settings are installed; no recorded run yet

"Real runs" means the installed instructions were executed by that agent's own CLI and the result was checked independently. Details per feature: platform matrix.

See agentic in action

A real agentic tui session, recorded key by key: pick Codex, Claude Code and OpenCode, add Context7, JEV and Telegram, choose DevOps infrastructure plus backend — then the same product-owner request runs in all three agents inside the freshly installed project. Every frame is captured from a real run.

Quick start

npx @jetrabbits/agentic@latest          # interactive installer

Or non-interactive, for all three verified agents:

npx @jetrabbits/agentic@latest install --project-dir . \
  --agent-os codex,claude,opencode --areas software \
  --specializations software.backend,software.general

Keep it current:

agentic upgrade            # new binary + refreshed project instructions, your edits preserved

Other install options (curl script, release binary, build from source): usage guide.

Your AI team

Installed with --profile team; the default lean profile adds only explorer and reviewer.

Role	What it does
🎯 product-owner	Turns a request into scope and acceptance criteria; accepts the result
🗺️ pm	Plans milestones and risks for larger work
🏗️ team-lead	Designs the approach and signs off on quality
💻 developer	Implements the change with tests
🔍 qa	Verifies the result and gives go / no-go
📐 analyst	Clarifies requirements, scenarios and API contracts
⚙️ devops-engineer	Handles CI/CD, infrastructure and deployment
🧹 instruction-reviewer · memory-curator	Review the run afterwards and suggest guidance fixes

Pick the specializations your project needs — 10 software (backend, frontend, mobile, data, ML, security, …) and 8 DevOps (Kubernetes, CI/CD, observability, SRE, …).

Honest numbers

We measure instead of promising. Latest round: 7 small tasks (bugfixes, a feature, a question, a scope trap, a bugfix in a file that already fails make lint, a multi-module bugfix and feature), 3 runs each, 21 runs per agent and mode. Cells show accepted runs and cost per accepted run (Codex and OpenCode: API-equivalent estimate).

Agent	Plain agent	Lean (default)	Lean + JEV	Role team + JEV
Claude Code (sonnet)	18/21 · $0.092	21/21 · $0.095	21/21 · $0.097	21/21 · $0.073¹
Codex (gpt-5.6-terra)	18/21 · $0.059	21/21 · $0.072	21/21 · $0.103	18/21 · $0.301
OpenCode (Qwen3.8-Flash-Next)	19/21 · $0.026	19/21 · $0.041	17/21 · $0.046	18/21 · $0.047

• The plain agent is the cheapest everywhere. Every plain failure is the lint task: it left make lint failing. Lean passed that task on every Claude Code and Codex run.
• Lean + JEV accepts as many runs as lean, or fewer, and costs 1.02–1.43 × lean. It stays opt-in.
• ¹ Prompt-cache effect: the role team's long fixed prefix was already cached by earlier runs.
• Not measured yet: large multi-step tasks. The multi-module tasks did not make the plain agent fail.

All numbers, per task: comparison.

Learn more

• Usage guide and installed CLI lifecycle
• Reference: layout, platform matrix, install options
• Claude Code integration · Telegram notifications
• Review pipeline · JEV hooks and real CLI testing
• Smart upgrade

Contributing

See CONTRIBUTING.md for authoring standards, templates and the pull request process.

License

See LICENSE if present, or contact the maintainers.