all-plan_skill

Overview

This skill enables collaborative planning across multiple mounted CLIs (Codex, Claude, Gemini, OpenCode) with Codex acting as the coordinator. It maintains persistent context and minimizes token overhead to support iterative solution design in real time. The goal is to produce coherent, multi-agent plans that combine strengths of different LLMs and tools.

How this skill works

Codex orchestrates a session that routes tasks and prompts to the best-suited CLI, gathers responses, and merges outputs into a single evolving plan. Persistent context is shared between agents so iterations build on prior work rather than repeating prompts, reducing token usage. The process supports refinement cycles where agents critique, expand, or implement parts of the plan until convergence.

When to use it

• Designing complex systems that benefit from multiple LLM perspectives (architecture, code, verification).
• Breaking large projects into coordinated tasks handled by specialized agents or tools.
• Generating, reviewing, and iterating on implementation plans with distributed expertise.
• Speeding up design sessions where real-time multi-agent feedback improves outcomes.
• Maintaining a persistent planning context across long-running or multi-step workflows.

Best practices

• Define clear, scoped tasks for each agent to avoid overlap and conflicting output.
• Establish a simple coordination protocol (task ownership, review steps, acceptance criteria).
• Keep context concise: summarize decisions and link to longer details only when needed.
• Use iterative checkpoints so Codex can reconcile agent outputs and enforce consistency.
• Monitor token use by trimming redundant history and storing stable conclusions externally.

Example use cases

• Architecting a microservices system: Claude proposes high-level tradeoffs, Gemini generates UML diagrams, OpenCode drafts API stubs, Codex coordinates and merges artifacts.
• Feature design sprint: agents brainstorm options, evaluate feasibility, produce implementation steps and test plans in a single coordinated session.
• Refactoring effort: Codex assigns modules to different agents for analysis, merges suggested changes, and creates an ordered rollout plan.
• Cross-platform CLI tool development where one agent focuses on UX, another on platform bindings, and others on tests and docs.