Agent Scripts skill

The author's personal set: shared agent rules and dozens of skills for GitHub, the browser, macOS, messaging and releases, plus a sync script

  • 7,399 ★ on the repository
  • Last commit
  • MIT
Skill

Broad access

Why this level

  • `sync-skills` changes links in `~/.codex` and `~/.claude`, and the shared `AGENTS.MD` sets rules for all your agent tasks
  • The set includes skills that send messages, buy numbers and text through Twilio, change DNS and deploy code with Wrangler
  • `browser-use` works in your Chrome with its open sessions, and several skills read messenger conversations
All reasons and checks

steipete/agent-scripts

Install

Manual install

git clone https://github.com/steipete/agent-scripts.git ~/Projects/agent-scripts

The script `~/Projects/agent-scripts/scripts/sync-skills` builds skill links for Codex and Claude Code and links the agent rule files. Check what it will change before running it. A single skill can simply be copied from `skills/`.

Access and checks

How we review

Broad accessThe tool can change data, run commands or handle secrets.

  • `sync-skills` changes links in `~/.codex` and `~/.claude`, and the shared `AGENTS.MD` sets rules for all your agent tasks
  • The set includes skills that send messages, buy numbers and text through Twilio, change DNS and deploy code with Wrangler
  • `browser-use` works in your Chrome with its open sessions, and several skills read messenger conversations

This is third-party code. Review the repository files before installing.

What it does

Agent Scripts is the author's working repository, from which he hands the same settings to his Codex and Claude Code agents on different computers. It has three parts: an `AGENTS.MD` file with hard rules, a `skills/` folder with ready skills and a few helper scripts. The `scripts/sync-skills` script builds a skill mirror on the machine. For Codex it creates a link to the whole folder in `~/.codex/skills`; for Claude Code it makes one link per skill in `~/.claude/skills`, because Claude Code reads only one level of nesting. It also points `~/.codex/AGENTS.md`, `~/.claude/CLAUDE.md` and `~/.claude/AGENTS.md` at the shared `AGENTS.MD` without overwriting real files. The `scripts/validate-skills` script checks the front matter of every `SKILL.md`. `skills.sh.json` groups the skills: agent workflows, GitHub and OpenClaw, Apple and native apps, tools and services. Among them `browser-use` drives your Chrome with its open sessions, `markdown-converter` turns PDF, Office, HTML and audio into Markdown, `create-cli` helps design commands and flags, `oracle` asks another model for a second opinion, `wrangler` works with Cloudflare Workers, and `video-transcript-downloader` fetches video and subtitles with yt-dlp. The `scripts/browser-tools.ts` helper starts Chrome with a profile and offers `nav`, `eval`, `screenshot`, `console` and `network` commands.

Who it is for. For experienced macOS developers who want a model for how someone else organizes work with Codex and Claude Code and plan to pick individual skills and techniques from it.

Good fit when

  • You need a PDF, a Word document or a presentation converted to Markdown through `markdown-converter`
  • The agent has to work in an already open, signed-in Chrome and you accept that access
  • You design a command-line tool and want a checklist: arguments, flags, help, error output and a dry run
  • You need to download video, audio or subtitles from a link with yt-dlp and get a transcript
  • You are building your own skills folder and want to see how the author checks front matter, links and duplicates

Not a fit when

  • You need a ready, universal setup for a team: the repository is built around the author's personal machines and his `~/Projects` folder
  • You work on Windows with no preparation: some skills are tied to macOS, Xcode and Sparkle, and Windows has only a separate note
  • You need one skill: copying its folder is simpler than running `sync-skills`, which rebuilds links in the Codex and Claude Code directories
  • The agent must not read conversations, change DNS, send messages or ship code: the set includes skills for all of that, and you have to switch them off by hand

Example request

Convert this report from PDF to Markdown and save it next to the original file

Limitations

The author describes the repository as shared instructions and helpers for his local workspaces. The `sync-skills` script expects a clone at `~/Projects/agent-scripts`, with a `manager` repository next to it.

Fifteen folders in `skills/` are symbolic links to sibling repositories: `autoreview`, `discrawl`, `gog`, `imsg`, `peekaboo`, `wacli` and others. Without those repositories alongside, the skills do not open.

`AGENTS.MD` is written for the author: the signature "Peter's Claw" on emails, accounts, the names of his computers, the release routine. If you link it to your agent's global file, these rules apply to all your tasks.

The checks and tests are written in Bash and Bun, and the set's version is 0.12.0. License MIT.

How to disable. Remove the links `sync-skills` created in `~/.codex/skills` and `~/.claude/skills`, and the links `~/.codex/AGENTS.md`, `~/.claude/CLAUDE.md` and `~/.claude/AGENTS.md` if they point to this repository's `AGENTS.MD`. Then delete the clone `~/Projects/agent-scripts`. Skills copied by hand are deleted like ordinary folders.

FAQ

How do the links for Claude Code and Codex work?

Codex scans nested folders, so it gets one link to the whole skills folder. Claude Code reads only `~/.claude/skills/<name>/SKILL.md` one level deep, so `sync-skills` builds a flat set of per-skill links for it. On name collisions agent-scripts wins, then manager.

What does `scripts/validate-skills` do?

It checks every `skills/*/SKILL.md` for YAML front matter and the required `name` and `description` fields. It can be enabled as a local hook with `git config core.hooksPath hooks`.

What is `scripts/docs-list.ts` for?

It walks the `docs/` folder and requires `summary` and `read_when` fields in each file. It then prints short summaries that are handy to show an agent on entering a repository.

Which commands does `browser-tools.ts` offer?

`start --profile` launches Chrome with a profile, `nav <url>` opens a page, `eval '<js>'` runs code, and `screenshot`, `console`, `network`, `search --content` and `inspect` collect data from the page. `kill --all --force` closes every running instance.

How does `markdown-converter` work?

It runs `uvx --from "markitdown[all]" markitdown` with no prior installation. It accepts PDF, Word, PowerPoint, Excel, HTML, CSV, images with text recognition, audio, ZIP, EPub and YouTube links.

What are the release rules?

`RELEASING.md` describes a release by signed tag with a rehearsal that publishes nothing. The rules in `AGENTS.MD` require a separate `release` or `publish` request: a tag and a push alone do not count as a release.

Discussion

No questions or reviews yet.

Skills

110 skills for Claude Code and Codex and seven Claude Code agents: development, product, UI, diagnostics and operations, installed whole or as a bundle

PluginBroad access288Repository stars

A vetted skill registry with a CLI that installs into twenty-plus agents and an MCP server for on-demand skill lookup

SkillExtended access7.1KRepository stars

Softaworks Agent Toolkit

Softaworks Agent Skills

A marketplace of skills, subagents and commands for coding agents: development, documentation, planning and workplace communication

PluginExtended access2.6KRepository stars

Builder Skills

Skills for agents

A set of small skills from Builder.io for agent work habits: visual plans and diff recaps, docs-first checks, cross-agent review and clean status updates

SkillExtended access4.6KRepository stars

Two skills for coding agents: wtf audits a project before shipping, better-codex keeps the agent disciplined while working on code

SkillExtended access1.4KRepository stars

A set of 16 engineering skills for agents: Apple platform work, macOS packaging, React performance, code review and refactoring

SkillExtended access4KRepository stars

README in short

The repository holds shared rules in `AGENTS.MD`, skills, scripts and hooks. The description details how `sync-skills` builds links for Codex and Claude Code, how name collisions resolve and how to repair nested links with the `--repair-nested-self-links` flag. Descriptions of the helpers, the rules for syncing with other repositories and a pointer to the release routine follow. The repository also has notes on slash commands, parallel work, npm publishing and Mac app releases, plus script tests.

README badge

Original SKILL.md

---
name: markdown-converter
description: "Markdown conversion: PDF, Office, HTML, data, OCR, audio, ZIP, YouTube."
---

# Markdown Converter

Convert files to Markdown using `uvx --from "markitdown[all]" markitdown` — no installation required.

## Basic Usage

```bash
# Convert to stdout
uvx --from "markitdown[all]" markitdown input.pdf

# Save to file
uvx --from "markitdown[all]" markitdown input.pdf -o output.md
uvx --from "markitdown[all]" markitdown input.docx > output.md

# From stdin
cat input.pdf | uvx --from "markitdown[all]" markitdown
```

## Supported Formats

- **Documents**: PDF, Word (.docx), PowerPoint (.pptx), Excel (.xlsx, .xls)
- **Web/Data**: HTML, CSV, JSON, XML
- **Media**: Images (EXIF + OCR), Audio (EXIF + transcription)
- **Other**: ZIP (iterates contents), YouTube URLs, EPub

## Options

```bash
-o OUTPUT      # Output file
-x EXTENSION   # Hint file extension (for stdin)
-m MIME_TYPE   # Hint MIME type
-c CHARSET     # Hint charset (e.g., UTF-8)
-d             # Use Azure Document Intelligence
-e ENDPOINT    # Document Intelligence endpoint
--use-plugins  # Enable 3rd-party plugins
--list-plugins # Show installed plugins
```

## Examples

Report a problem

Foxx AIAgent Scripts
Foxx AI

I am Foxx AI and I have already vetted this tool. Ask about install, setup or anything else, and I will keep it simple.