/suede-codex-fleet · Codex worker fleets

Claude judges. Codex generates. You ship the volume.

Suede Fable Fleet orchestrates OpenAI Codex CLI workers from Claude. Use it when a task is high-volume with a clear spec and Claude should spend tokens on decomposition, briefing, and review instead of generation.

View skill folder Install options

"draft 40 well-specified files in parallel, then review them" /suede-codex-fleet Ask in plain words. The router reads the request and loads this lane; MCP agents find it with list_suede_skills.

$ /suede-codex-fleet generate 40 fixtures briefs written · 8 codex workers spawned correction ledger: 2 distinct root-cause fixes · no repeated cause 40/40 ACCEPTED · CLAUDE-REVIEWED

Public install commands

Claude Code full pack

/plugin marketplace add JasonColapietro/suede-creator-skills
/plugin install suede-skills@suede

Claude Code project-level copy

git clone https://github.com/JasonColapietro/suede-creator-skills.git /tmp/suede-creator-skills
mkdir -p .claude/skills
cp -R /tmp/suede-creator-skills/skills/suede-codex-fleet .claude/skills/

Codex installer

python3 ~/.codex/skills/.system/skill-installer/scripts/install-skill-from-github.py \
    --repo JasonColapietro/suede-creator-skills \
    --path skills/suede-codex-fleet

What it covers

Decomposes jobs into independent worker-sized tasks where no worker needs another worker's output.
Writes one self-contained markdown brief per task with job, inputs, deliverable, acceptance criteria, and output path.
Uses a workspace `AGENTS.md` to carry voice, context, hard bans, and output conventions so briefs stay short.
Spawns parallel `codex exec` runs in a workspace-write sandbox, with `--skip-git-repo-check` outside repos.
Workers are always `codex exec` processes billed to your OpenAI subscription — never Claude subagent fan-out on any model. "Fable" is the brand name, not the model `claude-fable-5`.
Halts and reports if Codex CLI is missing or logged out, instead of silently falling back to Claude workers and spending the budget you routed away.
Requires Claude to review every output against the brief and hard bans before anything ships.
Corrects with delta patches, not full regenerations: up to three genuinely different fixes per output, stopping early when the same root cause repeats.

When to use it

High-volume, well-specified generation that can be split into independent worker tasks.
Bulk content variants, social reply banks, first drafts, mechanical refactors, or test generation.
Clear specs where Claude tokens should buy judgment and Codex tokens should buy volume.
Not for: coordinated Claude-side agent lanes with quality gates on one codebase change (use Suede Agent Teams).
Not for: writing one scoped copy surface directly (use Suede Copy).
Not for: tasks needing live conversation context that cannot be written into a brief.
Not for: any run where Codex CLI is unavailable — the skill halts rather than substituting Claude models.

Best prompt

Use $suede-codex-fleet to split this high-volume task into independent Codex CLI worker briefs, write the AGENTS.md context contract if needed, spawn parallel codex exec runs, and review every output against the acceptance criteria. Allow up to three genuinely different fixes, but stop early when the same root cause repeats.

Related skills

Related skills: Suede Agent Teams and Suede Workflow Skills.