Open Axis — Productivity skill for Claude Code

Open Axis

**A local control plane that routes engineering tasks across multiple AI CLIs (Claude Code, OpenCode, Codex) — instead of you manually deciding which tool, which model, and what to do when one fails.**

Built and used daily as the author's own multi-CLI dev environment. This is not a demo app — it is a working piece of infrastructure, documented here for review.

---

Overview

Modern AI-assisted development means juggling several CLIs (Claude Code, OpenCode, Codex/GPT), each with its own agents, skills, and models. Left manual, every task means re-deciding: *which tool, which model, what happens if it fails, and how do I not lose task state if I get interrupted.*

Open Axis is a Python control plane that sits **beside** those CLIs — not replacing them — and owns only the layer they don't provide natively: routing policy, capability tiers, fallback, durable task state, a cross-environment registry, and a security policy layer both CLIs read from.

Problem

AI development becomes inefficient when tasks are routed between tools by hand: no consistent policy for which model fits which task, no fallback when a provider fails mid-task, no shared record of what actually exists (installed agents/skills) across environments, and no durable state if a task is interrupted.

Solution

A single control plane, run locally, that classifies a task, selects the right skills/agent/model against real capability tiers, dispatches to whichever CLI backs that model, falls back deterministically on failure, and persists task state to SQLite so nothing is lost on interruption.

Architecture

TASK (user request)
      |
      v
+-------------------+   local, deterministic, no network/model call --
|   CLARITY GATE     |   skips triage entirely for obviously-clear tasks
+-------------------+   (asymmetric cost design: false "clear" is dangerous,
      |                 false "nee