Skip to main content

​
Overview

The @standardagents/skill package ships editable coding skills for Standard Agents work. Each skill is authored in plain .md files so teams can review, copy, and customize the guidance without changing a generated artifact. The first bundled skill is agentbuilder, which helps coding agents decide when to use:
  • defineModel
  • definePrompt
  • defineTool
  • defineAgent
  • defineHook
  • defineThreadEndpoint
  • subagents

​
Installation

This opens an interactive picker so you can choose the install target. Supported targets:
  • Codex and other coding agents (Cursor, Gemini CLI, GitHub Copilot, OpenCode, and others): installs the skill folder into ~/.agents/skills
  • Claude Code: installs the skill folder into ~/.claude/skills
Every target gets the same Agent Skills folder, agentbuilder/SKILL.md. Without a terminal the command installs into ~/.agents/skills. The skill tells your coding agent to read the published Standard Agent Specification and to treat it as correct when the skill and the specification disagree. There is no separate specification skill to install. Use --force to overwrite an existing installed copy:
Use --agent to skip the picker, and --dir to install somewhere else, such as a project-level .agents/skills folder:

​
Package Install

npm
pnpm
yarn

​
Package Layout

​
Helper API

Use the package API when you want to inspect or install the bundled skill from Node:

​
Manual Editing

The skill content is intentionally stored as raw markdown. A common workflow is:
Then edit:
  • .agents/skills/agentbuilder/SKILL.md

​
What The agentbuilder Skill Covers

  • Choosing the right Standard Agents Builder primitive for a change
  • Keeping prompt logic, tool logic, and orchestration concerns separate
  • Knowing when subagents are warranted
  • Mapping changes to the correct workspace files
  • Applying the docs-and-tests checklist before landing behavior changes
  • Reading the published specification before answering what it requires