Skip to main content
Don't Sleep On AI

Agent orchestration

Jev logoNow with Jev

Fable Foreman: Turn Claude Fable into your agent orchestrator

Opus 5.5 or Fable leads the work: it plans, routes each job to a capable model, and personally verifies the result before accepting it.

Download Fable Foreman
Jordan Olsen presenting Fable Foreman
Rev
v0.6.5
Tool
Fable Foreman
For
Claude Code
Platform
macOS + Windows
Rate
Free — members

Fable Foreman teaches Claude Opus 5.5 or Fable 5.1, the lead, how to plan your coding tasks, route each job to a capable model, and personally verify the result. Opus 5.5 is the current Claude lead and a qualified candidate for difficult review. The lead stays responsible for the outcome. Other models handle implementation, testing, and repairs.

Download Fable Foreman

Free with a Don't Sleep On AI account. Claude Code required. Codex and Grok are optional.

Checking your access…
Claude

DOWNLOAD FOR

Claude Code

Skill for Claude Code

⬇ Download Claude Code skill v0.6.4 (.zip)

Unzip it, then paste one line into Claude Code (below).

Now with Jev

Jev is optional structured decision support. It can help the lead sort what to look at first. It is not a coding or review replacement.

To use Jev, have an active OpenRouter account and API key with available balance. The included helper supports OpenRouter and direct TypeSafe. Jev is also available through Vercel AI Gateway, which needs a separate integration. Enable Jev locally if you want it. You opt in; the agent does not. Do not paste keys into chat. Ask your coding agent to store the key securely and run the access check.

The lead decides

Jev flags possible issues. The lead checks the evidence, keeps every finding, and accepts the result. The skill continues if Jev is off or unavailable.

What it helps you do

1

Compare models on one dated table The current lineup is Claude Opus 5.5 and Fable 5.1, OpenAI GPT-6 Astra, Sol, and Luna, and xAI Grok 4.7. They sit on one dated evidence table. Opus 5.5 is the current Claude lead for this skill and a qualified candidate for difficult review. Fable leads the same workflow.

2

See which model gets each job One command shows which model gets each kind of job, based on what is actually installed, signed in, and approved on your computer. Clear-cut jobs get a fixed default. Bounded factual or mechanical work can go to GPT-6 Luna, where the result is easy to verify. When two good options are close, such as which model writes everyday code, the lead asks you once, in plain language, recommends one, and remembers your answer for the rest of the session. An unattended run uses a sensible default and tells you what it chose.

3

Prove a provider can do the work A tiny test call per provider says whether Codex or Grok can really do work, not just whether you are signed in. The verdicts are plain: LIVE, SIGNED OUT, or BALANCE EXHAUSTED.

4

Say yes twice for expensive helpers GPT-6 Astra and Claude Fable as workers or reviewers need two approvals in the same session: first yes, then a confirmation that names the model and its cost. Choosing Fable or Opus as the lead does not use that rule, and starting the skill does not require two approvals.

5

Check the result before you accept it The real tests are run. An independent reviewer checks behavior-changing work. The usual reviewer is Claude Opus; you can name a different reviewer. When Claude built the change, a different AI family also reviews it when Codex or Grok is available. The lead then personally accepts the result and tells you what is complete and what still needs attention.

6

Tested on real Claude Code runs A behavioral test suite runs real Claude Code leads through 15 scenarios: farming work out, review by a different AI family, following your instructions, recovering when a provider runs out, and asking before expensive models. It grades their actual tool calls against hidden answer keys. Every scenario passed on its most recent run. Those were single runs, not a statistical rate. In a blind test, an AI model answered 12 routing questions using only the skill's instructions: with the previous version it got 6 right, and with v0.6 it got all 12.

How it works

1Describe the result you wantAsk for a feature, a bug fix, a refactor, or help planning a project. The lead identifies the work and how it will know the result is ready.
2Let the lead assign the workIt reads which model your machine will use and gives each agent a clear task, ownership, and checks. The lead keeps a working record of assignments, decisions, finished work, and open problems, and uses it for later choices. Small tasks stay simple, and independent work can run in parallel when useful.
3Review, repair, and verifyWorkers return their results and evidence. Repairs go back to the original builder when possible. Work that needs your input is recorded while independent work continues. The lead then personally accepts the finished work.
4Get a clear handoffSee what changed, how it was checked, and anything still unresolved. You keep control over publishing, deployment, and other actions that need your approval.

Start using Fable Foreman

1. Sign in with a free dontsleeponai.com account and download the ZIP.

2. Give the ZIP to Claude Code and ask it to install the Fable Foreman skill.

3. Start a new session and ask Claude to use /fable-foreman with your task.

If Claude Code needs the extracted-folder instruction
Install the Claude Code skill in ~/Downloads/fable-foreman: copy skills/fable-foreman to ~/.claude/skills/fable-foreman and every file in agents/ to ~/.claude/agents/. Confirm /fable-foreman appears in my skills.
Sign in to keep track of what you've copied

Try this

Use this at the start of a new session.

Try this
Use /fable-foreman to build the feature described in PLAN.md. Choose suitable agents, keep track of the work, and personally verify the finished result. Ask me before deploying.
Sign in to keep track of what you've copied

Before you start, these are the usual questions.

Other Foreman skills

Open another Foreman skill to view it and download.

Don't sleep on what's next. Free tools, prompts, and skills for Claude Code and OpenClaw — sent when they drop. No spam, unsubscribe anytime.

Give your next project a clear plan and a responsible lead.

Don't sleep on AI.

Changelog

0.6.52026-09-24Current

v0.6.5 — page copy update; skill still v0.6.4

This is a page-copy update. The downloadable Fable Foreman skill is still v0.6.4. The page now uses one Jev setup recommendation and states current lead, worker, and review roles in plain language.

0.6.42026-09-24

v0.6.4 — now with Jev

Fable Foreman now spells out Jev: optional structured decision support for the lead, with OpenRouter and TypeSafe on the included helper. Ask your agent to store the key securely and run the access check. The skill package itself is unchanged.

0.6.32026-09-24

v0.6.3 — smarter model routing

The lead compares Claude Opus 5.5, Fable 5.1, GPT-6 Astra, Sol, and Luna, and Grok 4.7 on one dated evidence table. Opus 5.5 is the recommended lead. One command on your machine shows which model gets each job. A small test call checks that a provider can really do work. GPT-6 Astra and Claude Fable are hired as helpers only after you say yes twice. Optional Jev can do sorting jobs on review findings with your own key, and the skill works the same without it. Every one of the 15 test scenarios passed on its most recent run. In a blind test, an AI model answered 12 routing questions using only the skill's instructions: with the previous version it got 6 right, and with v0.6 it got all 12.