OpenVibeCoding

The open command tower for AI engineering.

Stop babysitting AI coding work.

AI coding does not lack models. It lacks a command tower.

OpenVibeCoding is the command tower for AI engineering. It helps teams plan / delegate / track / resume / prove long-running engineering work across Codex and Claude Code instead of juggling scattered chats, local scripts, and after-the-fact logs.

OpenVibeCoding is a contract-first multi-agent orchestration repository.

The public story is intentionally narrower than the full monorepo:

• See one proven workflow first
• Choose the right adoption path second
• Open MCP / API / builder / skills surfaces only after the real job is clear

Current public boundary: OpenVibeCoding is a repo-backed command tower, not a hosted product, and the shipped MCP surface remains read-only.

Current lane order is deliberate:

• Primary lane = the read-only MCP package plus the Official MCP Registry entry
• Secondary lane = the adoption-router public skill packet
• Companion/example lane = local starter kits and coding-agent bundle examples, which are not the canonical public root

Quickstart · Watch the 18s Teaser · First Proven Workflow · Compatibility Matrix · Distribution Status · Public Docs · AI + MCP + API Surfaces · Builder Quickstart · Releases

If you want the fastest product read instead of the full distribution matrix, open the 18-second teaser first.

Why OpenVibeCoding Exists

Most agent demos stop at "the model replied." OpenVibeCoding is built for the next question: can we inspect what happened, review what changed, classify the workflow case, and rerun it without guessing?

The deeper product claim is straightforward:

• Plan the next move instead of improvising one more prompt.
• Delegate scoped work without losing the case record.
• Track live progress and queue posture from one command tower.
• Resume long-running work when humans step away or sessions degrade.
• Prove what happened with evidence, compare, replay, and approvals.

The engineering philosophy underneath that loop is equally explicit:

• Prompt Engineering: write the right worker brief, scope, constraints, and deliverables.
• Context Engineering: keep the right material in the right head, and treat explicit handoff as a fallback rather than the default loop.
• Harness Engineering: move work through contracts, runtime bindings, approvals, and proof surfaces so the system can keep operating safely.

OpenVibeCoding currently combines:

• Command Tower: one operator surface for governed AI work, live run visibility, queue posture, and L0-style oversight
• Workflow Cases: one stable operating record that ties request, verdict, proof, and linked runs together
• Proof & Replay: one place to inspect evidence bundles, compare reruns, and replay failures before promotion
• Control surfaces: a web dashboard plus a macOS desktop shell for the same command tower
• Read-only inspection surfaces: repo-local MCP, API, and contract read models that expose truth without turning mirrors into execution authority
• Governed boundaries: fail-clo

... View full README on GitHub