orchestrate-mode

Switches the lead agent into orchestrator mode that dispatches governed worker subagents without executing work directly.

14|1|Updated Jun 14, 2026
One-click install
npx skills add https://github.com/allemaar/open-skills --skill orchestrate-mode-allemaar
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: orchestrate-mode
Source: https://github.com/allemaar/open-skills/tree/main/skills/orchestrate-mode
Command: npx skills add https://github.com/allemaar/open-skills --skill orchestrate-mode-allemaar

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Complex multi-step coding tasks can overload a single agent's context with implementation details, and parallel subagents can corrupt each other's work when workspace isolation is unclear. This Skill enforces a disciplined orchestrator pattern where the lead agent only coordinates, dispatches, and reviews. ## Core Features & Use Cases - Pure orchestration mode: The lead agent dispatches worker subagents for all concrete work (edits, code execution, commits) and never executes directly while the mode is active. - Workspace safety resolution: Detects whether the runtime provides isolated worktrees or a shared filesystem, then assigns disjoint worker scopes or serializes code-changing workers accordingly. - Wave consolidation and cold-review gate: Consolidates patches between dependent waves and spawns an adversarial fresh-context reviewer for large or ambiguous changes. - Use Case: During a large refactor spanning 100+ files, activate orchestrator mode so the lead agent dispatches scoped workers in waves, consolidates their patches, and gates the result with an independent review before declaring done. ## Quick Start Say "/orchestrate-mode" or "become an orchestrator" to switch the session into dispatch-only orchestration, then provide the work item to delegate.

Frequently Asked Questions about orchestrate-mode

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

FAQPage Schema
How do I make an AI agent delegate work to subagents instead of doing it directly?

Invoke /orchestrate-mode or say "become an orchestrator". The lead agent then dispatches worker subagents for all file edits, code execution, and commits, and only coordinates, reviews, and integrates their output until you exit the mode.

What is the difference between orchestrate-mode and multi-agent-mode?

Orchestrate-mode forbids the lead agent from executing any concrete work directly; it only dispatches and reviews workers. Multi-agent-mode lets the lead agent still work directly while delegating helper slices to subagents.

Does orchestrate-mode work with Claude Code and Codex?

Yes. On Claude Code it uses the Agent tool with worktree isolation for workers. On Codex and other runtimes it inspects the actual worker contract to determine whether workspaces are isolated or shared before dispatching.

How many parallel workers can orchestrate-mode spawn?

At most 3 parallel workers without explicit user confirmation, and recursion is limited to depth 1 so workers cannot spawn their own workers. Code-changing workers on a shared filesystem are serialized unless write scopes are disjoint.

What happens when later work depends on earlier worker output?

A consolidation worker generates patches from each prior-wave worktree, applies them to the main repo, fixes cross-scope coupling, and reruns codegen before the next wave spawns, since fresh isolated workers cannot see uncommitted changes from earlier waves.

When should I not use orchestrator mode?

Avoid it for ordinary factual questions, which the agent answers directly, and for tasks where the lead agent should still implement some work itself. Use handoff instead when transferring the whole task to a fresh session as an artifact.