suede-codex-fleet

# Suede Fable Fleet The Suede Fable Fleet: a high-end Claude model is the admiral — it decomposes, briefs, and reviews — and parallel OpenAI Codex CLI workers are the fleet. Do not rename the folder or frontmatter `name` to match the brand. ## The workers are Codex processes — never Claude models **This is the economic point of the skill.** Codex workers bill to the user's OpenAI/ChatGPT subscription. Claude subagents bill to their Anthropic limit. Someone asking for a codex fleet is deliberately routing volume *off* the Anthropic meter — satisfying that request with Claude models inverts the cost model and spends the exact budget they were protecting. `codex exec` is therefore the only way a worker runs here. Never substitute `Agent`, `Task`, `Workflow`, subagent fan-out, or any other in-house orchestration for a worker — not with `fable`, not with `opus`, not with any model, not "just for this one batch". Claude's role is admiral only: decompose, brief, review, assemble. If you are about to spawn something that is not a `codex exec` process, you have left this skill — stop and re-read the routing table below. **Preflight failure is a halt, not a fallback.** If Codex CLI is missing, not logged in, or not on `PATH`, say which check failed and ask whether to proceed on Claude models, with a rough estimate of what that fan-out will consume. Never fall back silently. **What getting this wrong costs** (measured, 2026-07-27): a Claude-model fleet ran in place of an explicitl...