atelier-orchestrator

Orchestrates multi-agent discovery, planning, implementation, review, and distillation phases using runtime teams.

2|Updated Apr 29, 2026
One-click install
npx skills add https://github.com/siarhei-belavus/agent-public --skill atelier-orchestrator-siarhei-belavus
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: atelier-orchestrator
Source: https://github.com/siarhei-belavus/agent-public/tree/main/skills/atelier-workflow/atelier-orchestrator
Command: npx skills add https://github.com/siarhei-belavus/agent-public --skill atelier-orchestrator-siarhei-belavus

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve? Coordinating multi-agent software work (planning, implementation, review, distillation) often devolves into chaotic pane management, lost context, and premature reviewer activation. This Skill turns a lead agent into a disciplined workflow state machine that runs the full atelier lifecycle from one orchestrator window. ## Core Features & Use Cases - Phase-based runtime teams: Creates one ephemeral team per phase (plan, implement, distill) and shuts it down before the next, keeping panes minimal and state in packet artifacts. - Artifact-first reviewer activation: Spawns reviewers only after the artifact they review exists, and preserves planner/implementer live context across all review iterations. - Model-selection gate: Requires a human-approved provider/model/thinking-level table before any team creation or teammate spawning. - Initiative mode: Splits large efforts into child task packets across repos, hydrates children in dependency order, and syncs outcomes back to a parent initiative packet. - Use Case: A lead agent receives a feature request, runs discovery, spawns a planner then plan-reviewer, routes P# findings until approval, then runs an implementer with plan-satisfaction and maintainability reviewers until acceptance, and finishes with distillation. ## Quick Start Ask the agent to orchestrate the atelier workflow for your task, starting with discovery and a model-selection proposal before any planning begins.

Frequently Asked Questions about atelier-orchestrator

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I orchestrate a multi-agent planning and implementation workflow?▼

Run the lead as a state machine: do discovery first, create a planning runtime team, spawn the planner, then spawn the plan-reviewer only after a PLAN.md draft exists. Route findings back to the same live planner until approval, then repeat the pattern for implementation and distillation.

When should reviewers be spawned in an agent review loop?▼

Reviewers should start only after the artifact they review exists. The plan-reviewer starts after an initial PLAN.md draft, and code reviewers start only after the first implementation pass is committed and declared ready for review.

What is the difference between single-task and initiative mode?▼

Single-task mode runs one packet through plan, implement, review, and distill. Initiative mode frames a parent initiative, splits it into child task packets, hydrates children in dependency order, and syncs outcomes back to the parent before final distillation.

Why preserve the same planner and implementer across review iterations?▼

Keeping the same workers alive preserves their live context across draft and review cycles, so findings are routed back to an agent that already understands the work. Killing and respawning same-phase reviewers discards useful context.

Can the orchestrator write code or plans directly?▼

No. The orchestrator is the workflow state machine, not a worker. It owns phase transitions, task routing, and gate decisions, but planning, implementation, and reviews belong to the spawned specialist agents.