pragmatic-orchestration plugin
Skills and the porch CLI for agent orchestration: a strong model plans, cheaper ones code, several model families review.
- 66 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- Delegation mode launches other agents with full access to the environment and spends quotas of several subscriptions.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add plugins/pragmatic-orchestrationDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- Delegation mode launches other agents with full access to the environment and spends quotas of several subscriptions.
This is third-party code. Review the repository files before installing.
What it does
The repository ships three skills in the Agent Skills layout: `pragmatic-orchestration`, `coordinator-mode` and `remote-agents`. The main one turns any agent with skills into an orchestrator that calls external agents through the `porch` CLI. `porch` has five modes: independent code review by several model families at once, delegating a task to one exact agent with supervision and steering, searching local session histories, media work and reading remaining subscription quotas. Coordinator mode is built for days of work: it keeps a decision and plan journal in one file, survives context compaction and accepts executor results on evidence. Supported executors are Codex CLI, Claude Code, OpenCode, Grok Build, Devin CLI and Antigravity CLI.
Who it is for. Developers running multi-day tasks with several agents who want cheap models to implement while a strong one keeps decisions and acceptance.
Good fit when
- Code is reviewed independently by several models at once with no write access.
- Implementation goes to one chosen agent and the course is steered while it runs.
- One coordinator session leads a project for several days with a decision journal.
- An old decision must be found in the history of local agent sessions.
Not a fit when
- The task takes half an hour: orchestration adds more overhead than it saves.
- Executors must not get full access to the machine.
- A single agent is wanted without delegating to other vendors.
Example request
Enable coordinator-mode, create a journal for the task, hand implementation off through `porch delegate` and request review from two other model families.Limitations
The `delegate` mode runs an executor with full access to the environment, so it should not be used where such access is unacceptable. Subscriptions and quotas are spent at several vendors at once. Only macOS is fully field-tested, Linux and Windows 11 support is called experimental, and `review` and `delegate` on Windows need Git Bash. The model pairs the author recommends depend on current plans, and some come under temporary terms.
How to disable. Remove the plugin with `claude plugin uninstall pragmatic-orchestration` or delete the skill folders from the agent directory. Clean the `porch` configuration and journals separately.
FAQ
How do I install the skills?
Run `npx skills add CodeAlive-AI/pragmatic-orchestration@pragmatic-orchestration`, then the same command with `@coordinator-mode`.
What can porch do?
Five modes: `review`, `delegate`, `sessions`, `media` and `quota`.
Which executors are supported?
Codex CLI, Claude Code, OpenCode, Grok Build, Devin CLI, Antigravity CLI and, for review only, Gemini CLI.
How do I turn on coordinator mode?
Explicitly: with `/coordinator-mode` in Claude Code, or by calling the skill by name in other agents.
Is there a UI for monitoring?
Yes, the optional Porch UI: after a one-time build `porch ui --desktop` opens a window with runs and streaming output.
Discussion
No questions or reviews yet.
Related
PluginsOne agent launches others in the background: Claude Code, Codex, Copilot CLI, Grok Build and Pi, with a model chosen per stage.
A process kit for agents: a skill router, project memory in a DEV folder and result verification before saying done.
A skill that turns the main Claude Code session into a dispatcher: tasks go to background subagents and the chat never blocks.
A kit that gives OpenCode role-based agents, handoffs, independent review and task state that survives interruptions.
A set of skills that hand a task to a separate CLI agent while you review the diff and land the commit yourself
A bridge between local CLI agents: Codex, Cursor or Claude Code delegate work to Antigravity, Grok, Kimi and seven more.
README in short
The README describes the orchestration system, three skills, five `porch` modes, coordinator mode and recommended model pairs, then gives installation and a platform support table.