From Frontier
Orchestrate work with a frontier harness or model as planner, reviewer, and synthesizer while worker harnesses handle bounded research, coding, testing, and log reduction through adapters such as Pi and Codex. Use when the user wants Frontier, harness orchestration, provider-backed delegation, Pi, Codex, oMLX, Ollama, OpenAI-compatible models, or token-heavy codebase work coordinated through subagents.
How this skill is triggered — by the user, by Claude, or both
Slash command
/frontier:frontier-orchestrationThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Use the frontier orchestrator for judgment. Use worker harnesses for bounded
Use the frontier orchestrator for judgment. Use worker harnesses for bounded heavy lifting. The point is not to maximize the number of agents; it is to spend the strongest planning/review model on architecture, tradeoffs, integration, verification strategy, and final synthesis, while delegating token-heavy scans, narrow patches, test runs, and log reduction to configured models and providers.
/frontier:delegate (or the frontier-worker
subagent). It routes the packet to a configured model through a harness.Delegation goes through /frontier:delegate, which forwards to the
frontier-worker subagent. The worker is a thin forwarder over a deterministic
companion runtime that owns all harness mechanics. You choose the harness by the
shape of the task:
Each harness runs against the model/provider selected by backend configuration. That can be local or cloud. Provider mechanics live in the companion runtime, never in a prompt. Do not write harness command strings in this skill or in any worker prompt.
Every delegated task should be self-contained:
Good worker prompts are narrow enough that the worker can finish without asking for context from the parent conversation.
Tell workers to stop and report when:
Before presenting completion:
npx claudepluginhub welldundun/frontier --plugin frontierCreates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.