An open API service indexing awesome lists of open source software.

https://github.com/produtoramaxvision/claude-code-maxvision-orchestration

Dynamic skill+subagent orchestrator for Claude Code: reads CLAUDE-agents.md, audits local components, checks upstream versions, and proposes installs/updates from a curated catalog before dispatching the task.
https://github.com/produtoramaxvision/claude-code-maxvision-orchestration

claude-code claude-plugin claude-skills orchestration plugin-marketplace skill-router subagents version-check

Last synced: about 1 month ago
JSON representation

Dynamic skill+subagent orchestrator for Claude Code: reads CLAUDE-agents.md, audits local components, checks upstream versions, and proposes installs/updates from a curated catalog before dispatching the task.

Awesome Lists containing this project

README

          

> **Read in:** 🇺🇸 [English](README.md) | 🇧🇷 [Português (Brasil)](README.pt-BR.md)

# Claude Code MaxVision Orchestration

[![License: MIT](https://img.shields.io/badge/License-MIT-blue.svg)](LICENSE)
[![Conventional Commits](https://img.shields.io/badge/Conventional%20Commits-1.0.0-yellow.svg)](https://conventionalcommits.org)

> Dynamic skill + subagent orchestrator for Claude Code. Reads `CLAUDE-agents.md`, audits what is installed locally, checks upstream versions, and proposes installs/updates from a curated catalog before dispatching the task.

When you fire a complex task at Claude Code, this plugin makes sure the **right subagent** runs with the **right skills**, and that those skills are **fresh** — pulling missing pieces from a curated catalog of premium sources (Anthropic official, VoltAgent, Antigravity, Composio, Obra, Trail of Bits, shadcn, and more) only after explicit human approval.

---

## What it does

Given a task, the orchestrator:

1. Reads `~/.claude/CLAUDE-agents.md` and `~/.claude/CLAUDE-skills.md` (your global cheatsheets).
2. Inventories every skill and subagent currently installed (personal + plugin + project scopes).
3. Picks the best subagent and required skills for the task.
4. **Checks upstream versions** of every component it will use (1-hour cache, `--force-check` to bypass).
5. Identifies missing components and looks them up in a curated catalog (`skills/discover-skill/references/skill-sources.json`).
6. Presents **one consolidated plan** — updates + installs — and waits for `sim/all/skip-N/none`.
7. Executes approved actions in order (updates first, installs second), tracking origin in `.maxv-source.json` per skill.
8. Dispatches the subagent with the right skills loaded.

Nothing destructive runs without explicit user approval.

## Installation

### Option 1: via Marketplace (recommended)

```text
/plugin marketplace add produtoramaxvision/claude-code-maxvision-orchestration
/plugin install maxvision-orchestration@maxvision-orchestration
```

### Option 2: local dev (without installing)

```bash
git clone https://github.com/produtoramaxvision/claude-code-maxvision-orchestration.git
claude --plugin-dir ./claude-code-maxvision-orchestration
```

## Quick start

After installation, in any Claude Code session:

```text
/maxvision-orchestration:orchestrate refactor src/components/Foo.tsx to use shadcn Select and add tests
```

The plugin will:
- Identify that this needs `react-specialist` + `test-automator` subagents and `frontend-design` + `shadcn/ui` + `superpowers:test-driven-development` skills.
- Notice `shadcn/ui` is missing and `frontend-design` has a newer upstream sha.
- Show you a single approval prompt.
- After your `sim`, install/update and dispatch.

## Skills exposed by this plugin

| Skill | Purpose |
| --- | --- |
| `orchestrate` | Master entry point. Runs the 8-step orchestration. |
| `route-task` | Reads `CLAUDE-agents.md`/`CLAUDE-skills.md` and picks subagent + skills. |
| `discover-skill` | Looks up a component name/keyword against the curated catalog. |
| `check-version` | Compares local `gitCommitSha`/`commit_sha` with upstream. Cache-aware. |
| `update-component` | Applies an upstream update to a skill or agent (with confirmation). |
| `install-skill` | Installs a skill via one of three methods (sparse-clone, marketplace, npx antigravity). |
| `audit-local-skills` | Inventory + redundancy + budget report. |
| `refresh-catalog` | Re-fetches upstream metadata to refresh `skill-sources.json`. |

## Custom subagent

| Agent | Purpose |
| --- | --- |
| `orchestrator` | Coordinates the multi-step flow above. Permissioned for read + bash + Agent. |

## Catalog of sources

Documented in [docs/CATALOG-SOURCES.md](docs/CATALOG-SOURCES.md). 4 trust tiers:

- **Tier 1 (Official Anthropic):** `anthropics/skills`.
- **Tier 2 (Curated premium):** `VoltAgent/awesome-agent-skills`, `sickn33/antigravity-awesome-skills`, `obra/superpowers-skills`, `obra/superpowers-lab`, `ComposioHQ/awesome-claude-skills`, `jeremylongshore/claude-code-plugins-plus-skills`.
- **Tier 3 (Domain-specific):** `trailofbits/skills`, `expo/skills`, `shadcn/ui`, `lackeyjb/playwright-skill`, `chrisvoncsefalvay/claude-d3js-skill`, `K-Dense-AI/claude-scientific-skills`.
- **Tier 4 (Evaluate with caution):** `alirezarezvani/claude-skills` (high overlap with `knowledge-work-plugins`).

## Architecture

See [docs/ARCHITECTURE.md](docs/ARCHITECTURE.md) for the 8-phase flow diagram, cache layout (`~/.claude/cache/maxv-orchestration/version-check.json`), and `.maxv-source.json` schema.

## Adding a new source to the catalog

See [docs/EXTENDING.md](docs/EXTENDING.md).

## CLI helper

The plugin ships a bash CLI at `bin/maxv-skills`, automatically added to PATH while the plugin is enabled. Use it outside Claude Code sessions:

```bash
maxv-skills list # list all local skills
maxv-skills find # search catalog + local
maxv-skills check # version check (uses cache)
maxv-skills check --force # bypass cache, re-fetch upstream
maxv-skills install # interactive install with confirmation
maxv-skills update # interactive update
maxv-skills audit # redundancy + budget report
```

## Requirements

- Claude Code ≥ 2.0
- `gh` CLI authenticated (`gh auth status`) — used by `check-version`
- `jq` available in PATH — used by version comparison
- `npx` (Node ≥ 18) — only required for the antigravity install path

## Hard rules (guardrails)

- Never installs or updates silently. Single approval prompt with `sim/all/skip-N/none`.
- Never dispatches to `general-purpose` Agent (denied by user policy).
- Max 3-4 subagents in parallel.
- Updates run before installs (updates can change resolution of new installs).
- Persists install method and origin in `.maxv-source.json` per skill folder for repeatable updates.

## Attribution

This plugin orchestrates third-party skill collections under their respective licenses (MIT/Apache-2.0). See [NOTICE](NOTICE) for full attribution and [LICENSE](LICENSE) for plugin-level terms.

## Contributing

See [CONTRIBUTING.md](CONTRIBUTING.md). PRs welcome for:
- New entries in `skill-sources.json` (source verification required)
- Improvements to the version-check strategy for non-git sources
- Bug fixes in the install methods

## Security

Report vulnerabilities privately via the process in [SECURITY.md](SECURITY.md). Do not open public issues for security concerns.

## License

MIT — see [LICENSE](LICENSE).

© 2026 Produtora MaxVision