Skip to content
Back to skills

Orchestration Kitty

ASecurity

Parallel execution using multiple Claude instances in Kitty terminal

  • 16 stars
  • 0 votes
  • 0 copies
  • 1 view
  • Added May 26, 2026
ai-agentsrustbashrefactoringgitapi

Works with

  • terminal
  • cli
  • api

Security analysis

A100/100

Scanned May 27, 2026

npx -y skills add Roberdan/convergio-community --skill orchestration-kitty --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Orchestration Kitty?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Orchestration Kitty
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/roberdan-orchestration-kitty/badge)](https://www.skillsdirectory.com/skills/roberdan-orchestration-kitty)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: orchestration
description: Parallel execution using multiple Claude instances in Kitty terminal
allowed-tools:
  - Read
  - Glob
  - Grep
  - Bash
  - Task
context: fork
user-invocable: true
version: "1.0.0"
---

# Multi-Worker Parallel Orchestration Skill

## Overview

Parallel execution of plans using Claude and/or Copilot CLI workers in Kitty terminal tabs. Supports mixed-engine mode where Claude handles complex tasks and Copilot handles mechanical ones.

## When to Use

- Complex tasks with 4+ independent subtasks
- Release preparations with multiple verification steps
- Large refactoring across multiple file domains
- Any plan created by `strategic-planner` agent

## Requirements

| Requirement | Details                                               |
| ----------- | ----------------------------------------------------- |
| Terminal    | **Kitty only** (not Warp/iTerm/Terminal.app)          |
| Config      | `allow_remote_control yes` in kitty.conf              |
| Claude      | `wildClaude='claude --dangerously-skip-permissions'`  |
| Copilot     | `copilot` CLI + `GH_TOKEN` or `COPILOT_TOKEN` env var |
| Max Workers | 4 (hard limit)                                        |

## Quick Start

```bash
# Claude-only (default):
orchestrate.sh <plan> 4

# Copilot-only (all workers use copilot --yolo):
orchestrate.sh <plan> 4 --engine copilot

# Mixed (Claude for complex, Copilot for codex:true tasks):
orchestrate.sh <plan> 4 --engine mixed

# Monitor (detects both Claude-N and Copilot-N tabs):
claude-monitor.sh

# Launch single worker by type:
worker-launch.sh claude "Claude-2" <task_db_id> --cwd /worktree
worker-launch.sh copilot "Copilot-3" <task_db_id> --cwd /worktree

# Standalone Copilot worker (no Kitty needed):
copilot-worker.sh <task_db_id> --model claude-sonnet-4-5 --timeout 600
```

## Integration with Agents

### strategic-planner

The `strategic-planner` agent can create plans with Claude assignments and execute them in parallel:

```
@strategic-planner Create an execution plan for [task] with parallel execution
```

When asked "Vuoi eseguire in parallelo?", it will:

1. Verify Kitty environment
2. Launch Claude workers
3. Send tasks to each worker
4. Monitor progress
5. Report completion

## Plan Format

Plans for parallel execution must include:

```markdown
## 🎭 RUOLI CLAUDE

| Claude   | Role        | Tasks      | Files           |
| -------- | ----------- | ---------- | --------------- |
| CLAUDE 1 | Coordinator | Monitor    | -               |
| CLAUDE 2 | Implementer | T-01, T-02 | src/api/        |
| CLAUDE 3 | Implementer | T-03, T-04 | src/components/ |
| CLAUDE 4 | Implementer | T-05       | src/lib/        |
```

## Critical Rules

1. **NO FILE OVERLAP** - Avoid git conflicts
2. **MAX 4 WORKERS** - Beyond = chaos
3. **VERIFY LAST** - lint/typecheck/build at end
4. **GIT COORDINATION** - One commit at a time
5. **WORKTREE MANDATORY** - Every worker runs `worktree-guard.sh` first. NEVER on main.
6. **COPILOT: --yolo** - Always use `copilot --yolo` for full autonomy (no confirmation prompts)

## Copilot CLI Flags Reference

| Flag               | Purpose                                         |
| ------------------ | ----------------------------------------------- |
| `--yolo`           | Full autonomy, no confirmations (REQUIRED)      |
| `--add-dir <path>` | Trust worktree directory                        |
| `--model <model>`  | Select model (claude-sonnet-4-5, gpt-4.1, etc.) |
| `-p "prompt"`      | Non-interactive mode (for scripted execution)   |

## Scripts

| Script                   | Purpose                                             |
| ------------------------ | --------------------------------------------------- |
| `orchestrate.sh`         | Main orchestrator (--engine claude\|copilot\|mixed) |
| `worker-launch.sh`       | Launch single worker in Kitty tab                   |
| `copilot-worker.sh`      | Standalone Copilot task execution                   |
| `copilot-task-prompt.sh` | Generate Copilot-compatible prompt from DB task     |
| `worktree-guard.sh`      | Block execution if not in correct worktree          |
| `claude-monitor.sh`      | Monitor all worker tabs (Claude + Copilot)          |
| `kitty-check.sh`         | Verify Kitty setup                                  |

## Related

- Agent: `.claude/agents/core_utility/strategic-planner.md`
- Global config: `${CLAUDE_HOME:-.claude}/commands/planner.md`
- Drift check: `plan-db.sh drift-check <plan_id>`

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.