Skip to main content

Agent orchestration

Sol Foreman: Orchestrate Your Build

Let GPT-6.1 Sol brainstorm, plan, and orchestrate your other Codex, Claude, or Grok agents. Luna handles bounded factual or mechanical work that is easy to verify.

Download Sol Foreman
Sol as the Sol Foreman orchestrator
Rev
v0.6.0
Job
Sol Foreman
For
Codex
Platform
Codex
Rate
Free — members

I built Sol Foreman so GPT-6.1 Sol can lead a Codex crew. Sol can lead, build, and review. It oversees other agents, assigns work, arranges independent reviews, and verifies that the work meets your requirements. GPT-6 Luna handles bounded factual or mechanical work that is easy to verify. This does not guarantee lower cost.

Download Sol Foreman

Free for members. It's a zip you hand straight to Codex.

Checking your access…
Codex

DOWNLOAD FOR

Codex CLI

Skill for Codex CLI

⬇ Download Codex CLI skill v0.6.0 (.zip)

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

Now with Jev

Jev is optional structured decision support. It can help Sol narrow prioritization and triage. It is not a coding or review replacement. When you turn Jev on, Sol uses it by default to merge review findings, check whether cited evidence supports a finding, and screen forwarded text for instructions aimed at the agent. Sol asks for the newest Jev release (Jev 1.14 today).

To use Jev, have an active OpenRouter account and API key with available balance. The included helper supports OpenRouter. 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 agent to help configure OpenRouter and verify that a Jev request succeeds.

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.

Install it

Download the zip, unzip it, then paste one line into Codex.

1Download the zipLog in with a free member account, tick the disclaimer, and download Sol Foreman.
2Unzip itExtract the zip. It contains one top-level sol-foreman folder.
3Paste this into CodexHand Codex the unzipped folder and this line.Install this unzipped sol-foreman folder globally at ~/.agents/skills/sol-foreman. Read its references/compatibility.md first: reconcile active runs, preserve local changes, and stage the new version before replacing anything. Archive older global copies, including ~/.codex/skills/sol-foreman, outside skill-discovery folders without deleting their source checkout. Confirm v0.6.0 is the only globally discoverable Sol Foreman skill.
4You're done whenOpen a new Codex session. You're done when $sol-foreman resolves as a skill instead of plain text.

Use it

Paste this above your actual request.

Paste into Codex
Use $sol-foreman to lead this task toward a finished outcome. Define the proof first. Delegate substantive implementation, checks, ordinary repairs, and authorized delivery to qualified workers unless the whole delivery path is cheaper for Sol to handle directly. Keep Sol responsible for planning, routing, independent review when required, and personal verification of the assembled result.
Sign in to keep track of what you've copied

What it does

1

Keeps the work pointed at finished It keeps the original unfinished outcome visible, names the evidence that would prove it, and sets a stopping point before it starts spending time on a route.

2

Makes the task fit the route Sol can lead, build, and review. It gives qualified workers coherent work through authorized delivery, including checks and ordinary repairs. Sol works directly when that costs less across the whole delivery path, and revisits the choice if the work grows. Savings are not guaranteed.

3

Uses the newest models Sol finds the newest release in each model family your account can use: today GPT-6.1 Sol rather than the older GPT-6 Sol. A newer release is used after one quick test call; a brand-new kind of model is researched first.

4

Expensive models only when you ask Codex Astra (any version) and Claude Fable are never picked automatically, never used as a backup, and never suggested as an option. Sol uses one only if you ask for it and then confirm again once it names the model and its cost.

5

Adapts to the crew you have GPT-6 Luna handles bounded factual or mechanical work that is easy to verify. Codex-only is a complete way to run it. Claude and Grok are optional when they are available and you want to use them; your stated preferences and an observed limit can change the route, but they do not lower the quality bar.

6

Learns from the last attempt It keeps a short written record of the outcome, route, result, and review so the next choice can respond to actual quality evidence instead of a vague claim that a model is better or cheaper.

7

Bounds lead, worker, and review progress It asks for a first useful artifact and finite evidence checks. If the same outcome misses twice, the lead must change the plan, take over, or name a real external blocker instead of sending another automatic loop.

Good to know

I designed Sol to run with Codex alone. Claude and Grok are optional routes, not requirements. If an optional provider reaches a known limit, Sol reports that fact and reroutes eligible work instead of waiting for a reset.

Still deciding? 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 Codex, Claude Code, and OpenClaw — sent when they drop. No spam, unsubscribe anytime.

An agent that says "done" has only told you it stopped. Make it prove the rest.

Don't sleep on AI.

Changelog

0.6.02026-10-22Current

v0.6.0 — newest models, expensive helpers only when you ask

Sol now finds the newest model in each family by itself: today that is GPT-6.1 Sol rather than the older GPT-6 Sol. A newer release is used after one quick test call, and a brand-new kind of model is researched first. Sol never hands work to Codex Astra (any version) or Claude Fable on its own; it uses one only if you ask for it and then confirm again once it names the model and its cost. When Jev is turned on, Sol uses it by default to merge review findings, check cited evidence, and screen forwarded text, and it asks for the newest Jev release (Jev 1.14 today). Real GPT-6.1 Sol leads ran 13 test scenarios, and all passed on their latest run after fixes to the scoring checks. Those were single runs, not a success rate.

0.5.02026-09-24

v0.5.0 — GPT-6 Sol lead, Luna, optional Jev

GPT-6 Sol can lead, build, and review. GPT-6 Luna handles bounded factual or mechanical work that is easy to verify. Optional Jev can narrow prioritization. Savings are not guaranteed. Astra and Sol honor user restrictions; they do not use Fable/Opus Foreman's two-yes helper rule.

0.4.12026-09-05

v0.4.1 — qualified workers carry ordinary delivery

I updated Sol Foreman so Sol can keep qualified workers responsible for implementation, checks, ordinary repairs, and authorized delivery. Sol still plans the work, ensures independent review, and personally verifies the assembled result before accepting it.