# Orchestrate Multi-Model Skill Cascades

Complex AI work often needs several skills, models, and quality checks. This skill helps design cascades with routing, sandbox testing, swarm coordination, and memory.

## Install

```bash
npx skillstore add dnyoussef/cascade-orchestrator
```

## Metadata

- Status: approved
- Slug: dnyoussef-cascade-orchestrator
- Version: 2.0.0
- Author version: 2.0.0
- Skillstore revision: r1
- Version status: valid
- Tree hash: af91cfe18f67a8cb8a08e17ba5193195ef14ad08838a7a11a8e98c3755773fee
- Author: DNYoussef
- GitHub username: DNYoussef
- License: MIT
- Repository: https://github.com/DNYoussef/ai-chrome-extension/tree/main/.claude/skills/cascade-orchestrator
- Ref: c579efd02184f1128d6741d16e6c8d951d7d4208
- Supported tools: Claude, Codex, Claude Code
- Audit status: complete
- Agent install advisory: allowed
- Manual install advisory: allowed
- Artifact signature: available
- Audit attestation: unavailable
- Human verification: not\_verified
- Risk factors: external\_commands
- Quality score: 81
- Quality tier: silver
- Public page: https://skillstore.pages.dev/skills/dnyoussef-cascade-orchestrator
- Manifest: https://skillstore.pages.dev/api/skills/dnyoussef-cascade-orchestrator/manifest

## Capabilities

- Designs sequential, parallel, conditional, codex-sandbox, multi-model, and swarm-parallel stages.
- Routes work across Claude, Codex, and Gemini-style models based on task requirements.
- Defines Codex sandbox loops with tests, auto-fix attempts, and escalation limits.
- Adds shared memory fields for reading and writing cascade state across stages.
- Produces structured cascade plans with inputs, stages, error handling, and integration settings.
- Documents workflow examples for development, modernization, and bug fixing.

## Use Cases

- Plan development quality pipelines: Create staged workflows for research, implementation, testing, review, and pull request reporting.
- Route work across AI models: Assign large-context, search, media, rapid prototype, and reasoning tasks to suitable model routes.
- Design test-and-fix audits: Map quality gates that run tests, attempt sandbox fixes, recheck regressions, and escalate failures.

## Prompt Templates

### Create a basic cascade

```
Create a cascade for my goal: [goal]. Use these micro-skills: [skills]. Include stages, inputs, outputs, and error handling.
```

### Add model routing

```
Design a cascade for [project]. Route stages across Claude, Codex, and Gemini-style models based on context size, speed, search, and reasoning needs.
```

### Build a test-and-fix workflow

```
Create a Codex sandbox cascade for [repository or feature]. Include test stages, auto-fix iterations, regression checks, and escalation criteria.
```

### Design production orchestration

```
Design a production cascade for [system]. Include swarm parallelism, shared memory, GitHub integration, model routing, quality gates, and failure recovery.
```

## Limitations

- It provides workflow designs and examples; it does not execute cascades by itself.
- Named micro-skills and model routes must exist in the user environment.
- GitHub, ruv-swarm, and memory steps require separate tooling and permissions.
- Auto-fix behavior still needs human review before production changes are merged.

## Best Practices

- Define clear stage inputs, outputs, and ownership before adding parallel execution.
- Set explicit retry, rollback, and escalation rules for every automated fix stage.
- Use memory only for necessary context and name each stored key clearly.

## Anti Patterns

- Do not route every task to the same model when stages have different needs.
- Do not enable auto-fix loops without tests, iteration limits, and human review.
- Do not reference micro-skills, MCP tools, or GitHub actions that are unavailable.

## Security Audit

- Audited at: 2026-07-05T12:41:48.932\+00:00
- Summary: All static findings were false positives caused by Markdown fence delimiters, illustrative YAML or Python examples, and model-routing prose. The reviewed skill is documentation-only and shows no executable helper code, unauthorized network access, secret access, data exfiltration intent, or prompt-injection attempt.

## Stats

- Views: 215
- Downloads: 10
- Favorites: 0
- Popularity score: 0