* Implement pack registry system and management CLI Add registry.json manifests to agent packs and commands to support structured discovery. Include a CLI for project initialization, a registry generator script, and a plugin version locker. Add a new learn command for session documentation. * Add CLI and centralized workflow registry Implement an interactive CLI tool to browse and install workflow packs. Consolidate decentralized manifests into a single root registry for easier maintenance. * update * buffer * Rewrite CLI as TUI workflow selector Replace the Commander-based CLI with a TUI built on Bun, Solid.js, and OpenTUI. This version provides hierarchical navigation for selecting workflow packs and standalone items. - Migrate code from packages/cli to howaboua-cli - Support global (~/.config/opencode/) and local (.opencode/) sync - Add one-shot verification rules to howaboua-cli/AGENTS.md * chore: ignore @howaboua-cli directory --------- Co-authored-by: OpenCode Agent <agent@opencode.ai>
Vite + React 19 + TypeScript 5.9 + Convex + Tailwind 4.1 Agents
A specialized collection of Opencode agents for building modern full-stack web applications with the Vite + React 19 + TypeScript 5.9 + Convex + Tailwind 4.1 stack.
Stack Overview
- Vite: Fast build tool and development server
- React 19: Latest React with Server Components, Actions, and Compiler
- TypeScript 5.9: Modern TypeScript with latest features and strict defaults
- Convex: Reactive database and serverless backend
- Tailwind CSS 4.1: Utility-first CSS framework with new v4 features
Note: this will also work with Bun. Just make sure to initialise your project with Bun rather than Vite.
Available Agents
Core Technology Agents
VRTCT-orchestrator.md- Master coordinator for the full stack, orchestrating subagents (does not write code)VRTCT-brain.md- Stack knowledge base and implementation lead (writes code)convex-database-expert.md- Deep Convex expertise for schema design, queries, mutations, and backend patternsreact-19-master.md- React 19.2 mastery including Server Components, Actions, and the Component Engineering Specification (viacomponent-engineeringskill)typescript-59-engineer.md- TypeScript 5.9 expertise with advanced typing and modern patternstailwind-41-architect.md- Tailwind CSS 4.1 expertise including new v4 features and theming
Reference Documentation
CODING-TS.md- Universal engineering guidelines for TypeScript projectsCONVEX.md- Comprehensive Convex development guide and best practicesREACT19.md- Complete React 19+ reference with modern patternsTS59.MD- TypeScript 5.9+ language reference and configurationTAILWIND4.md- Tailwind CSS 4.1 complete reference and migration guide
Specialized Skills
The pack includes five specialized Convex skills for deep backend implementation:
convex-core- Base data modeling, functions, and validation patterns.convex-runtime- HTTP actions, file storage, search indexes, and scheduling.convex-auth- Authentication (Clerk/WorkOS/Auth0) and Convex Auth patterns.convex-deploy- Deployment workflows, CI/CD, and environment management.convex-components- Isolated backend components (Agent, RAG, Workpool, Workflow).
Spec-Driven Component Engineering
This stack incorporates a formal Component Engineering Specification via the component-engineering skill.
Commands
/component-review [file]: Audits a React component against a11y, composition, and styling pillars./component-create [name] [intent]: Generates professional, spec-compliant UI artifacts.
Key Principles
- Semantic First: Uses native elements for built-in a11y.
- asChild Pattern: Composable APIs via Radix Slot.
- Data-Driven Styling: Stable
data-slotanddata-statehooks for Tailwind v4. - Artifact Taxonomy: Standard classification (Primitive, Component, Block).
Usage
These agents are designed for:
- Global installation in
~/.config/opencode/agent/for reuse across projects - Project-specific installation in
.opencode/agent/directories - Specialized development when working with this specific tech stack
Each agent includes YAML frontmatter with usage guidance, mode constraints, and when to invoke the agent. Agent instructions use RFC 2119 keywords and XML tags for unambiguous behavior.
By default, VRTCT-orchestrator.md is set up to only be the primary agent (coordination only, no code), while VRTCT-brain.md handles implementation. The rest can be used either as main agents or subagents. This separation prevents your orchestrator from getting bogged down in implementation details.
Scoped Skills and Isolation
Important
To prevent context pollution and ensure predictable behavior, VRTCT agents follow the Principle of Least Privilege. Every agent MUST have all irrelevant skills disabled, only enabling the specific skills required for its domain.
convex-database-expert.md: Only allowsconvex-core,convex-runtime,convex-auth,convex-deploy, andconvex-components.VRTCT-brain.md: Only allowscomponent-engineering.- Others: Explicitly set to
"*": "deny"to ensure zero external skill interference.
Recommended changes to frontmatters
Select your preferred model for the agents. I highly recommend putting a high-context model as the vite-react-convex-expert.md agent, since this will be your main subagent orchestrator. Theoretically it should only delegate tasks and verify the outputs, so by using a big context model your session can go for a while without interruption. Consider Gemini 3 Pro for this. Opus or a GPT model for Convex. Gemini 3 Pro knows a bit of Tailwind 4.1, so it's a good contender. GLM is surprisingly good for frontend. The "cheapstack" is the free tier Gemini CLI and GLM for subagents.
(...)
</commentary>
</example>
mode: primary
model: google/gemini-3-pro-preview
---
(...)
Recommended AGENTS.md inclusions
To avoid the agents constantly trying to start new Vite/Bun/Convex processes, you should include the following in your AGENTS.md file:
Vite/Bun/Convex will be running at all times. You are to NEVER start a new process for these. Your main interactions with them should be through linters and convex codegen command.
