Agent Scripts

Agent Scripts is an open-source repository by Peter Steinberger that provides shared agent instructions, reusable workflow skills, and dependency-light helper scripts for use with Codex/Claude-style AI coding agents. It is available on GitHub under the MIT License and is designed to be used across multiple local workspaces and downstream repositories.

What It Is

Agent Scripts is a personal-but-shareable toolkit that acts as the canonical source of truth for AI agent configuration across projects. It centralizes hard rules for agents (via AGENTS.MD), reusable workflow skills (in skills/), portable helper scripts (in scripts/), and local guardrails (in hooks/). The repository is structured so that global agent tools like Codex and Claude can symlink to it, ensuring consistent behavior across all projects.

How the Skills System Works

Skills are the primary routing layer in Agent Scripts. Each skill lives in skills/<name>/SKILL.md and uses YAML front matter with a name and a short description field optimized for agent routing rather than documentation. Key conventions include:

• Descriptions are kept short and generic to aid routing.
• Repeatable workflow commands are placed in skills/<name>/scripts/.
• Skills can be validated with scripts/validate-skills.
• Public shared skills from the OpenClaw ecosystem and repo-owned skills are exposed via tracked relative symlinks (e.g., skills/autoreview, skills/discrawl, skills/birdclaw).

Helper Scripts

The repository includes several dependency-light utility scripts:

• scripts/committer: Stages listed files, enforces non-empty commit messages, and runs skill validation before committing.
• scripts/validate-skills: Validates YAML front matter and required fields in every skills/*/SKILL.md; can be enabled as a git hook.
• scripts/docs-list.ts: Walks a docs/ directory, enforces summary and read_when front matter, and prints onboarding summaries.
• scripts/browser-tools.ts: A standalone Chrome DevTools helper supporting commands like start, nav, eval, screenshot, console, network, search, content, inspect, and kill. Can be compiled to a binary with Bun.

Deployment and Sync Model

Agent Scripts is designed to be symlinked globally so that tools like Codex and Claude automatically pick up shared rules:

• ~/.codex/skills and ~/.claude/skills point to the skills/ directory.
• ~/.codex/AGENTS.md, ~/.claude/CLAUDE.md, and ~/.claude/AGENTS.md all point to AGENTS.MD.

Downstream repositories use a pointer-style AGENTS.MD that references the shared file rather than copying its contents, keeping rules centralized and easy to update. When syncing, the repo is treated as canonical and downstream repos pull the latest before applying changes.

OpenClaw Ecosystem Connection

The repository is part of Peter Steinberger's broader OpenClaw ecosystem of open agent tools. Public OpenClaw shared skills live in a sibling agent-skills repository and are exposed in Agent Scripts via symlinks. Steinberger announced in February 2026 that he is joining OpenAI and that OpenClaw will move to a foundation to remain open and independent.