Will maximum compression always lower my session cost?

Not automatically. Total cost includes instructions, repo context, diffs, tool output, logs, clarifications, and retries. Measure the same task at the same quality bar, since terse output that causes rework is not cheaper.

Should I paste the whole skill file into AGENTS.md?

Do not. Keep stable repo instructions in AGENTS.md, specialist workflows in skills, and the current objective in the prompt. Codex discovers repo skills without duplication.

How do I pick a Caveman mode without risking risky work?

Choose by misunderstanding cost: normal for architecture and handoffs, lite for discovery, full for implementation loops, ultra only for narrow mechanical work with strong verification.

Are the install commands guaranteed to stay current?

The guide lists install commands for project-local or global skill setup and global Caveman install, but product defaults, platforms, and telemetry can change upstream.

What files does the repository actually contain?

A README, a complete web-friendly Markdown guide, a 14-page PDF setup guide, a token-conscious AGENTS.md template, and Codex prompts and session workflows.

What four concerns does the stack separate?

AGENTS.md for stable project rules, the TypeSafe AI skill for typed semantic judgments, Caveman for session-level response conciseness, and the task prompt for the current outcome and constraints.

Is this official OpenAI, TypeSafe or Caveman documentation?

No. It is an independent community reference, and it warns that commands, defaults, platforms, licensing and product behavior may change, so review upstream docs before installing.

CommunityJev Agents & MCP

Lean Codex Stack

Reusable Codex setup separating stable AGENTS.md rules, the TypeSafe skill, Caveman conciseness modes, and the task prompt, with a Markdown guide and printable PDF.

See highlights

Open github.com

Stars: 1
Published by: Community
Author: KelvinxG
Last verified: Oct 2026

KelvinxG/lean-codex-stack

User repository on GitHub

View on GitHub

A practical guide to TypeSafe AI, Caveman modes, AGENTS.md, and token-efficient Codex workflows

Forks: 0
Last pushed: Updated Sep 2026

Repo stats from the GitHub API, cached Oct 2026.

About this project

A practical setup for TypeSafe AI, Caveman, AGENTS.md, and token-conscious Codex workflows. Read the complete Markdown guide · Download the PDF. Lean Codex Stack is a reusable reference for organizing AI-assisted development without putting every rule, workflow, and preference into one oversized prompt. It explains how to separate four concerns: TypeSafe AI installation and usage guidance; practical Codex prompts and session workflows; and caveats for measuring token savings honestly.

Quoted from the project's README. Read the full README.

Tagscommunity github

Works forJev for Tool calling

Used inJev for Software & DevTools

Highlights

Four-layer split keeps project rules, semantic judgments, response conciseness, and task scope in separate places.
Install guidance covers project-local and global TypeSafe skill setup plus global Caveman install and mode switching.
Mode table maps normal, lite, full, and ultra to suitable work and warns ultra away from migrations and security.
Honest-numbers section warns short answers can cost more once context, logs, clarifications, and retries are counted.

Watch out

Independent community reference, not official OpenAI, TypeSafe, or Caveman documentation; verify commands and behavior upstream first. Copy reviewed against the repository README on 2026-10-04, not from running it.

Questions about this entry

Short answers drawn from the linked project or page.

More Jev resources like this

jev-mcp

MCP server that exposes Jev classify, score, yes-no check, risk gate, and multi-question decide tools to coding agents through one Jev call per tool.