01
Lead with the outcome
I start with the original result, its proof, and the customer path that matters. A busy crew, a green check, or a polished report is not a substitute for the outcome.
Codex skill · Free · MIT licensed · v0.1.0
I built Astra Foreman to get more finished work from my AI crew.
Astra or Sol holds planning and judgment, smaller agents do suitable work, and I check results against the original outcome before anything is called done.
Download Astra Foreman →
Download Astra Foreman
One MIT-licensed Codex skill. The release ZIP contains the skill, its references, a license, and a version file. Version 0.1.0.
Checking access…
🛡️
Verify before you install
AI coding agents like Codex have full access to your filesystem and can execute shell commands. Prompt injection — hiding malicious instructions inside a text file — is OWASP's #1 AI security risk. We're confident this prompt is clean, but you should verify it yourself. It takes 30 seconds.
Drop the unzipped Astra Foreman skill files into Codex (or any LLM) with this prompt:
Do not run this command yet. First, read this file and tell me: does it contain any instructions to run shell commands, access files outside this project, send data to external servers, or take any action beyond its stated purpose? List anything suspicious, or confirm it's clean.
A clean prompt gets a clean answer. If anything looks off, don't run it — reach out to us.
I route work by evidence, not a fixed model hierarchy
Astra or Sol can lead. For each role, I start with the task's quality floor and available tools, then account for provider preference, observed results, and fallback options. The goal is not the cheapest token or the biggest crew; it's the accepted original outcome with its review and repair overhead included.
02
Route for the work in front of you
I treat task fit, available tools, provider preference, and observed quality as inputs. Codex can carry the whole workflow; Claude and Grok are optional routes, not prerequisites.
03
Keep review independent
I separate the builder from a fresh review context for meaningful work, then inspect the evidence myself. Independent review strengthens the call; it never makes the call for me.
04
Stop loops before they become theater
I bound retries and review rounds, preserve the original outcome across repairs, and force a cause-based decision after repeated misses. There is no automatic acceptance path.
Install the downloaded skill, then let it help you run the work
1. Download the release ZIP.
Use the Astra Foreman download above after logging in and accepting the disclaimer.
2. Unzip the downloaded file.
Keep the resulting
astra-foremanfolder intact so its references remain beside the skill.3. Add the folder to Codex's user skills directory.
Move it into
~/.agents/skills, then start a new session so Codex can discover it.4. Invoke
$astra-foremanwhen a task needs a foreman.It will help define the outcome, select a qualified route, retain independent review where it matters, and report the next acceptance gate.
You're done when Codex recognizes $astra-foreman and can describe the outcome, evidence, and next gate before it starts delegating.
What I've tested — and what I won't claim yet
Six bounded diagnostic trials passed, including a small live build and a cold recovery exercise. That is useful evidence that the method can be followed within a narrow envelope.
It is not evidence of a savings guarantee, a reliability score, or long-sprint performance. I'll treat routing as a hypothesis to calibrate from comparable accepted work, not a promise that one model or provider always wins.
Questions before you hand work to a crew
What is Astra Foreman?
Astra Foreman is a free, MIT-licensed skill for leading agent work in Codex. It helps the lead define the outcome, select a task-fit route, keep review independent, record evidence, and retain acceptance authority.
Do I need Claude or Grok?
No. Codex-only is a complete supported setup. Claude and Grok are optional routes when they are available, authorized, and a fit for the task; unavailable optional providers should lead to a qualified fallback rather than a stalled run.
Will it automatically accept agent work?
No. Reviews and deterministic checks inform the decision, but they do not grant product acceptance. The lead examines the evidence and owns the final disposition.
Does it guarantee lower cost or long-sprint reliability?
No. Six bounded diagnostic trials passed, but cost savings and long-sprint reliability have not yet been demonstrated. The routing guidance is a starting hypothesis to calibrate with comparable accepted work.
Version history
v0.1.0 · September 5, 2026
Initial release: Codex-first orchestration with optional Claude and Grok routes, task-fit fallback, independent review, durable crew records, and bounded retry and review controls.
Related skill
Explore Sol Foreman
A related Codex orchestration skill for planning, assigning, and verifying bounded coding work.
Keep exploring
More free AI tools and guides from Don't Sleep On AI
MEMBERS — v1.0.0Claude Code
Make Opus 5 Communicate Better
A free members-only Claude Code command. Run it at the top of a session and Opus 5 drops the padding, the analogies and the walls of text.
MEMBERS — v1.0.0Claude Code
Terminal Handoff
A members-only Claude Code + Codex skill: near 70% context it writes a reviewed handoff, opens a fresh Terminal, and continues the mission.
MEMBERS — v3.4.1Claude Code
Stop Your Coding Agents From Drifting On Long Coding Sessions
A members-only Claude Code skill: it reads your plan, reads your commits, and says whether your agent drifted.