Efficient Codex Execution

English | 简体中文

Less agent theatre. More finished work.

Efficient Codex Execution is a lightweight Agent Skill for a familiar frustration: a capable coding agent can still turn a small task into a ceremony—asking for permission to make routine edits, restating constraints you already gave it, or explaining an approach you explicitly ruled out.

This skill turns those instructions into execution behavior: complete routine reversible work, keep implementation-only constraints quiet, and surface only real decisions, risks, blockers, and evidence.

The problem, at a glance

When you say	Unhelpful agent behaviour	With this skill
“Rename this module and update imports.”	“Would you like me to proceed?”	Rename it, update imports, run the relevant check, report the result.
“Do not use LoRA.”	A paragraph explaining why LoRA will not be used.	Exclude LoRA silently and choose an appropriate alternative.
“Keep the public API unchanged.”	Repeats the rule in every progress update.	Preserve the API and call out only a meaningful compatibility concern.

What it teaches an agent to do

• Act on ordinary, reversible implementation work without creating a needless approval loop.
• Treat excluded approaches as silent solution-space constraints, not a topic for repeated commentary.
• Keep routine checks internal; communicate the result when it matters.
• Finish with concise, useful evidence: what changed, what was checked, and what still needs attention.

Fast by default. Never reckless.

This is not a shortcut around real judgment. The skill does not ask an agent to bypass required approval, hide security issues, dismiss destructive actions, or guess through material ambiguity. It removes noise so those moments stand out.

Get started

Install as a standalone skill

For runtimes that support Agent Skills, copy skills/efficient-codex-execution into that runtime’s skills directory. A common cross-runtime location is:

~/.agents/skills/efficient-codex-execution/

The copied folder must contain SKILL.md at its root.

Use as a Codex plugin

This repository is also a Codex plugin package. Its plugin manifest discovers the bundled skill under skills/, so you can add the repository through the local-plugin or marketplace workflow you already use.

Apply it to one project instead

Do not need a global skill? Copy the contents of:

skills/efficient-codex-execution/templates/AGENTS.md.snippet.md

into your project’s AGENTS.md.

What’s included

• skills/efficient-codex-execution/SKILL.md — the reusable skill.
• skills/efficient-codex-execution/tests/pressure-scenarios.md — behavioural pressure tests.
• skills/efficient-codex-execution/templates/AGENTS.md.snippet.md — a compact project-level version.
• .codex-plugin/plugin.json — the Codex plugin manifest.

A concrete example

Prompt:

Refactor this module. Do not use LoRA. Keep the public API unchanged.

Desired behaviour:

1. Refactor the module.
2. Preserve the API.
3. Silently exclude LoRA from the solution space.
4. Mention LoRA only if you later ask why it was excluded.

The aim is not a quieter agent for its own sake. It is an agent whose attention stays on the work—and whose messages earn their place.