Run local multi-agent work sessions with Orkas
Use Orkas to coordinate local-first multi-agent desktop work sessions across built-in specialist agents and installed coding CLIs such as Claude Code, Codex, OpenCode, OpenClaw, and Hermes.
npx skills add agentskillexchange/skills --skill run-local-multi-agent-work-sessions-with-orkas
Use Orkas when an operator wants one local desktop control plane for multi-step work that needs several specialist agents or local coding CLIs.
What this skill actually does
The repeatable workflow is to install Orkas, connect bring-your-own model keys, choose or configure specialist agents, and give the Commander a goal such as research, code work, document production, deck creation, UI design, or file automation. The Commander plans the work, delegates steps to specialist agents, runs supported local CLI agents as sessions when needed, and returns files and intermediate work on the user’s machine.
Invoke this instead of using a single chatbot or one coding CLI directly when the task benefits from coordinated specialist roles, a shared plan, local files, and cross-tool handoff. The scope boundary is the operator workflow of supervising a local multi-agent work session from Orkas; it is not a generic listing for desktop AI apps, LLM providers, or agent frameworks.
Inputs and prerequisites: Orkas desktop app, local workspace access, bring-your-own model API keys, optional installed coding CLIs such as Claude Code, Codex, OpenCode, OpenClaw, or Hermes.
Setup notes: Download the Orkas desktop app for macOS, Windows, or Linux from the upstream project site or GitHub releases, configure model provider keys, then use the Commander to plan and delegate work across built-in agents and optional local CLI agents.
Source and verification boundary: use https://orkas.ai?source=gh-orkas as the canonical reference before running the workflow; keep commands, API calls, CLI usage, and generated outputs reviewable against that upstream source.
Framework fit: publish this as a Multi-Framework workflow only when the operator can invoke the documented toolchain directly, rather than treating the upstream project as a generic product listing.