Orchestra

Get an independent read-only second opinion at a commitment boundary, before substantive work on a hard task, when stuck or changing approach, or at a final review gate. Use for "consult the advisor", "get a second opinion", "ask codex", "ask Astra", "ask gpt-6-astra", "ask Fable", "ask opencode", or "ask a bigger model". The advisor returns guidance; the main session keeps execution and decision ownership.

Run the Claude Code harness on GPT-5.6 Sol through a local CLIProxyAPI proxy when Anthropic usage runs out, and diagnose that setup when it drifts. Use for "my Anthropic usage ran out", "keep working on another model", "run Claude Code on GPT-5.6 Sol", "set up claudex", "claudex isn't working", or "bill against my Codex subscription". macOS + Homebrew.

Route bounded implementation from a capable main session to cheaper workers while keeping planning, review, verification, and git in the main seat. Use for worker dispatch, model arbitrage, parallel implementation, Sol, Luna, Astra, Muse, Grok, OpenCode, or native workflows.

Design or harden a software factory: an agentic loop that iterates toward a goal with a verification gate, persistent state, and a stop condition. Use for "build a loop", "agentic loop", "self-iterating agent", "/loop", "/goal", "Ralph loop", "maker-checker", "agentic SDLC", "ADW", picking a verification gate, bounding a loop's blast radius, unintelligible auto-merged PR titles, or loop process dumped into a GitHub PR body.

Dispatch many subagents in coordinated waves with per-wave review. Use for "fan out agents", "wave dispatch", "batch agents", "generate N variations", or any fan-out beyond about five parallel subagents that needs batching, ordering, and result reconciliation.

This skill should be used when the user says "deploy a team", "spin up agents to work on this", "use all our agents", "coordinate specialists", or wants to break a large task into parallel sub-tasks handled by multiple domain experts simultaneously. Orchestrates Claude Code's experimental agent team system using the full installed specialist roster.

This skill should be used when the user asks to "design the workflow visually", "show me the workflow before running it", "let me configure the agents first", "visual workflow builder", "which models for which steps", "let me pick the models", "plan this fan-out", "diagram the orchestration", or wants to review and adjust a multi-agent job — models, agents, phases, isolation — before it runs. Renders an editable graph (nodes, labeled edges, reject-back gates) the user can rewire; staffing is on the selected card. Emits a paste-back spec from the live graph. Builds on the coordinator skill; use coordinator alone when no visual review is wanted.

This skill should be used when opening a GitHub pull request, running `gh pr create`, writing a PR title or body, editing a PR description, or when a PR dumps loop state (NEXT CYCLE STARTS HERE, checker verdict, ticket-id prefix).