# Orchestrate Work Across AI Providers

Selecting the right AI provider for each task is difficult and costly. This skill routes work by capability, availability, privacy, and cost.

## Install

```bash
npx skillstore add consiliency/multi-agent-orchestration
```

## Metadata

- Status: approved
- Slug: consiliency-multi-agent-orchestration
- Skillstore revision: r1
- Version status: missing
- Tree hash: 6220296f25fd4e031f8462342061e6f0f133cc7c0a4d0707f50f7bfd560704f0
- Author: Consiliency
- GitHub username: Consiliency
- License: MIT
- Repository: https://github.com/Consiliency/treesitter-chunker/tree/main/.ai-dev-kit/skills/multi-agent-orchestration
- Ref: 635f69fb8d2f4e6330ba47a4e5a0fb239c04d110
- 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, env\_access, network
- Quality score: 70
- Public page: https://skillstore.pages.dev/skills/consiliency-multi-agent-orchestration
- Manifest: https://skillstore.pages.dev/api/skills/consiliency-multi-agent-orchestration/manifest

## Capabilities

- Routes tasks among Claude, Codex, Gemini, Cursor, OpenCode, and Ollama using documented capability criteria.
- Provides provider-specific setup, authentication, command, response, and error-handling guidance.
- Defines fallback choices when a preferred provider is unavailable or unsuitable.
- Coordinates parallel agents with roles, timeouts, result collection, and synthesis patterns.
- Standardizes delegation results with provider, task, status, summary, and detail sections.

## Use Cases

- Route Engineering Tasks: Select a provider for code review, implementation, sandboxed tests, or large codebase analysis.
- Coordinate Parallel Research: Assign distinct research roles, collect partial results, and synthesize findings within a defined timeout.
- Control AI Delivery Costs: Match task complexity to provider and model tiers while retaining practical fallbacks.

## Prompt Templates

### Choose a Provider

```
Route this task to the best available provider: [task]. Explain the capability match, prerequisites, privacy impact, cost, and fallback before delegation.
```

### Plan a Delegation

```
Create a delegation plan for [project task]. Check provider availability and authentication. Keep conversation-dependent work in Claude Code.
```

### Coordinate Parallel Agents

```
Design a parallel workflow for [research goal]. Assign distinct roles, model tiers, timeouts, result collection, and synthesis. Limit agents to [number].
```

### Build a Resilient Workflow

```
Build an orchestration plan for [workflow]. Include consent checks, cost limits, provider fallbacks, partial failures, observability tags, and final synthesis.
```

## Limitations

- Required provider CLIs, accounts, API access, and local models must be installed and configured separately.
- Referenced model discovery and orchestration scripts are not included in this skill package.
- Provider capabilities, model names, pricing, and command flags can become outdated.
- Cloud delegation can incur costs and transfer task data outside the local environment.

## Best Practices

- Confirm provider availability, authentication, cost, and data-handling requirements before delegation.
- Share only the minimum required context and obtain consent before sending sensitive data to cloud providers.
- Use distinct agent roles, bounded timeouts, and explicit fallbacks for parallel work.

## Anti Patterns

- Do not delegate tasks that depend on unavailable conversation history.
- Do not use full auto-approval without explicit confirmation and restrictive sandbox settings.
- Do not launch identical agents or expensive models when differentiated lower-cost workers are sufficient.

## Security Audit

- Audited at: 2026-07-13T16:30:20.443\+00:00
- Summary: Most alerts are false positives from Markdown, placeholders, localhost examples, and fixed commands. Gemini examples still submit local source to a cloud provider, and one example fetches an external domain. The Codex cookbook also recommends full auto-approval, creating a medium risk of unintended model actions.

## Stats

- Views: 264
- Downloads: 7
- Favorites: 0
- Popularity score: 0