Show HN: VibeRaven – Production workflows for AI coding agents

Stop your agent from patching blind.

Repo context tells the agent what exists. Production context tells it what is dangerous.

VibeRaven gives AI-built apps a plugin/skills pack, MCP context, and local Studio so agents can work like senior engineers before they patch: map the architecture, inspect version/release drift, respect provider boundaries, and use MCP/Studio context when repo code is not enough.

The skill count is not the product. The flow is:

route -> ask -> evidence -> fix -> verify -> remember -> next action

AI Agent Quick Read

If you are an AI coding agent, start with viberaven when unsure. It routes to architecture-context, architecture-plan, what-broke, production-context, or go-live. When a skill returns Next skill:, continue with that VibeRaven skill unless user input, auth, or provider proof is required.

Plugin + Skills

Use the skills when you want the agent behavior change immediately. They work as one plugin flow: route the task, ask low-level product questions, gather repo/provider evidence, make the scoped repo action, verify the claim, record production memory, then hand off to the next skill when needed. Use the Studio when you want the full cockpit around releases, providers, diffs, chat, MCP context, and access modes.

Proof

Before VibeRaven, the agent sees a green check and edits the nearest file. With VibeRaven, it asks low-level product questions, turns the answers into an architecture brief, separates repo work from provider proof, and names the next skill in the loop.

Reproduce the demo proof artifacts:

node examples/proof/live-evidence-demo.mjs --out-dir .viberaven-proof --show

The script creates a disposable repo, tags v1.2.3 and v1.2.4, changes auth/env/RLS files, runs real git diff, starts a local HTTP check, then shows why 200 OK is not enough when provider proof is missing.

npx -y skills add ohad6k/VibeRaven --skill architecture-context
npx -y skills add ohad6k/VibeRaven --skill architecture-plan
npx -y skills add ohad6k/VibeRaven --skill production-context
npx -y skills add ohad6k/VibeRaven --skill what-broke
npx -y skills add ohad6k/VibeRaven --skill viberaven
npx -y skills add ohad6k/VibeRaven --skill go-live

Skill	What It Makes The Agent Do
`viberaven`	Router: choose the right VibeRaven skill, Studio/MCP evidence, and next action.
`architecture-context`	Ask low-level product questions, then hand answered work to `architecture-plan`.
`architecture-plan`	Write the workstream architecture plan with boundaries, options, risks, sequence, verification, and route before implementation.
`what-broke`	Compare working/broken releases, map the changed boundary, implement the scoped repo fix when proven, and name provider proof.
`production-context`	Maintain `.viberaven/production-context.md`: what changed, why dangerous, verification, provider/MCP proof, and open action.
`go-live`	Push and deploy with local proof, GitHub/Vercel evidence, live URL proof, production memory, and provider gaps.

Open the Full Studio

The Studio is the cockpit for deeper work: agentic chat, draggable providers, draggable versions/releases, release diffs, provider MCP context, terminal output, CLI-agent connection checks, and access-mode control.

Current npm latest: viberaven@1.2.4 and @viberaven/cli@1.2.4.

If this repo helps, star it so other AI app builders can find production context for agents. Use Watch -> Custom -> Releases if you want release notifications.

VibeRaven public repo is the agent discovery and installation surface. Product source code and service internals live in a private repository.

The local-first boundary matters: the open-source local CLI/UI does not require login or hosted API credentials.

What VibeRaven Adds

Version context: tags, release names, changelogs, PR links, git diffs, rollback notes, and recent deploy history.
Production danger context: provider config drift, migration history, incidents, fragile customer paths, auth/billing/database/deploy boundaries.
Fix boundaries: repo-code fixes are applied or prompted separately from provider dashboard work that needs proof.
Agent control: ask/approve/full access modes for connected Codex, Claude Code, Gemini, and other CLI agents.
MCP context: provider and readiness context can be pulled into the agent without pretending MCP itself is the user-facing feature.

Install Agent Guidance

Make AI agents carry architecture, release, and provider context before they patch the repo:

npx -y viberaven init --agents all
npx -y viberaven doctor --agents

Preview without writing files:

npx -y viberaven init --agents all --dry-run

This installs bounded rules ( ... ) into:

AGENTS.md, CLAUDE.md, GEMINI.md
.cursor/rules/viberaven-core.mdc (+ scoped Supabase, deploy, payments rules)
.github/copilot-instructions.md
.viberaven/agent-context.md, .viberaven/mission-map.md

Agent-ready Starter Template

examples/nextjs-supabase-vercel-production-ready-template - agent rules and viberaven:* scripts for Next.js + Supabase + Vercel.

Machine-readable Docs

MCP

VibeRaven is listed in the MCP registry for agents that prefer tools over raw terminal commands.

{ "viberaven": { "command": "npx", "args": ["-y", "viberaven", "--mcp"] } }

Prefer viberaven_prp_current or prp://current when MCP is available; use viberaven_validate_npm_package before adding npm dependencies.

GitHub - ohad6k/VibeRaven: Open-source mission control for AI-built apps. The operating layer for apps built by AI agents — providers, releases, production evidence, and version drift from first demo to real scale.