# Agent Ecosystem Generator

> Claude Code plugin that scaffolds a validated multi-agent ecosystem (orchestrator + N sub-agents + M skills/MCPs) from a natural-language brief. A step up from skill-creator — with semantic validation, trigger-fire evals, and custom MCP scaffolding.

- **Type:** MCP server
- **Install:** `agentstack add mcp-prgilabert-agent-ecosystem-generator`
- **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior
- **Seller:** [prgilabert](https://agentstack.voostack.com/s/prgilabert)
- **Installs:** 0
- **Category:** [AI & ML](https://agentstack.voostack.com/c/ai-and-ml)
- **Latest version:** 0.1.0
- **License:** MIT
- **Upstream author:** [prgilabert](https://github.com/prgilabert)
- **Source:** https://github.com/prgilabert/agent-ecosystem-generator

## Install

```sh
agentstack add mcp-prgilabert-agent-ecosystem-generator
```

Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent.

## About

# agent-ecosystem-generator

> **Status: alpha.** Core flow smoke-tested end-to-end on a synthetic brief. Project-mode and custom-MCP paths are present but less exercised. Issues and PRs welcome.

Claude Code plugin that scaffolds a validated **orchestrator + N sub-agents + M skills/MCPs** ecosystem from a brief — a step up from `skill-creator`, which only scaffolds a single skill.

## What it does

1. `/generate-ecosystem` spawns a **`requirements-interviewer`** sub-agent that talks to you in isolated context and produces a `spec.json`.
2. The orchestrator picks an orchestration pattern (orchestrator-workers, sequential pipeline, routing, parallelization-voting, evaluator-optimizer), enumerates agents/skills/MCPs, and writes `plan.md`.
3. You pick the output target: **portable plugin** or **in-project `.claude/` config**.
4. A **`ecosystem-builder`** sub-agent materializes the files (+ scaffolds custom stdio MCPs in Python/FastMCP or TS SDK when needed).
5. A **`ecosystem-validator`** sub-agent runs semantic validation (overlap detection, pushy-ness scoring, coverage, MCP wiring) and trigger evals adapted from `skill-creator`. If anything fails, it loops back with fix hints up to 3 rounds.

## Install

Inside a Claude Code session, register this repo as a marketplace and install the plugin:

```
/plugin marketplace add prgilabert/agent-ecosystem-generator
/plugin install agent-ecosystem-generator@prgilabert-plugins
```

Or, if you've cloned the repo locally:

```
/plugin marketplace add "/absolute/path/to/agent-ecosystem-generator"
/plugin install agent-ecosystem-generator@prgilabert-plugins
```

> `claude plugin install ` (the CLI subcommand) resolves names through a registered marketplace — you can't point it at a local directory directly. The two-step flow above is the canonical path for local plugin repos until Claude Code ships first-class local installs.

## Usage

```
/generate-ecosystem review PRs with three checkers: security, performance, tests
```

Or just describe what you want — the `ecosystem-generator` skill auto-triggers on phrases like "design a multi-agent system", "build an agent team", "scaffold an orchestrator".

## Examples

See [`examples/`](./examples) for real ecosystems produced by this plugin. Each one is an installable Claude Code plugin you can inspect or `claude plugin install` directly.

- [`examples/pr-review-trio/`](./examples/pr-review-trio) — orchestrator-workers pattern: `/review-pr` orchestrates 3 parallel reviewers (security, performance, tests), aggregates via the `pr-comment-formatter` skill, references the `github` MCP. *Synthetic reference.*
- [`examples/space-dialogue-plugin/`](./examples/space-dialogue-plugin) — orchestrator-workers with parallel fan-out: `/escribe-dialogo` runs a plot writer first, then astronaut-voice + robot-voice in parallel, then a dialogue-reviewer that assembles and saves a Markdown file. *Generated by a real `/generate-ecosystem` session (Spanish brief, 4 iterations, 0.95 validation).*

## Improvements over `skill-creator`

| Feature | `skill-creator` | `agent-ecosystem-generator` |
|---|---|---|
| Single skill vs ecosystem | Single | Orchestrator + N agents + M skills + MCPs |
| Output targets | Skill dir | Portable plugin **or** `.claude/` project config |
| Custom MCP scaffolding | Referenced only | FastMCP (Python) **and** `@modelcontextprotocol/sdk` (TS) templates |
| Validation depth | YAML fields present | Overlap detection, pushy-ness scoring, coverage, MCP wiring, trigger evals |
| Orchestration pattern pick | N/A | Pattern library with justification (5 canonical patterns) |
| Context isolation | N/A | Interviewer + builder + validator all run in isolated sub-agent ctx |

See [stack_explanation.md](stack_explanation.md) §1–6 for the background, and [ROADMAP.md](ROADMAP.md) for what's next and where to contribute.

## Source & license

This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.

- **Author:** [prgilabert](https://github.com/prgilabert)
- **Source:** [prgilabert/agent-ecosystem-generator](https://github.com/prgilabert/agent-ecosystem-generator)
- **License:** MIT

Install and usage instructions live in the source repository linked above.

## Pricing

- **Free** — Free

## Security capabilities

Automated source analysis of v0.1.0 — what this tool can access:

- **Network access:** no
- **Filesystem access:** no
- **Shell / process execution:** no
- **Environment & secrets:** no
- **Dynamic code execution:** no

*"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.*


## Versions

- **0.1.0** — security scan: passed — Imported from the upstream source.

## Links

- Listing page: https://agentstack.voostack.com/l/mcp-prgilabert-agent-ecosystem-generator
- Seller: https://agentstack.voostack.com/s/prgilabert
- Browse the marketplace: https://agentstack.voostack.com/browse

---
Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.