agent-skills

Battle-tested skills for disciplined AI-assisted development.
Because your AI agent should have good habits, not just good vibes.

---

The Problem

AI coding agents are powerful but undisciplined. They forget what they were doing after context compaction. They write "updated stuff" commit messages. They start five things at once and finish none. They skip code review because "it's just a small change."

These skills fix that.

Skills

Skill	What it does
tracking-tasks	Never lose work to compaction — enforces task tracking across context boundaries
exploring-in-parallel	Stop searching sequentially — launch multiple subagents simultaneously
preserving-context	Capture working state before compaction so you can resume seamlessly
committing-code	Conventional commits with gitmoji — no more "updated stuff"
managing-agents-md	Create and maintain AGENTS.md so any AI agent can navigate your codebase
addressing-pr-feedback	Fetch, organize, and address GitHub PR review comments systematically
reviewing-code	Catch N+1 queries, security issues, and missing tests before they ship
writing-skills	Meta-skill — how to write skills that actually trigger and work
organizing-files	Organize macOS files across Desktop, Documents, Downloads, and iCloud Drive

Quick Start

Claude Code (recommended)

# Step 1: Add the marketplace
/plugin marketplace add oryanmoshe/agent-skills

# Step 2: Install the plugin
/plugin install agent-skills@oryanmoshe-skills

Done. All 9 skills are now available in every project.

Claude Code — Manual

git clone https://github.com/oryanmoshe/agent-skills.git
cp -r agent-skills/skills/* ~/.claude/skills/

Cursor

git clone https://github.com/oryanmoshe/agent-skills.git
cp -r agent-skills/skills/* .cursor/skills/

GitHub Copilot

cp -r agent-skills/skills/* .github/skills/

Other Tools

These skills follow the Agent Skills open standard. Any compatible tool can use them — just copy the skill folders to your tool's skills directory.

How Skills Work

Each skill is a SKILL.md file with YAML frontmatter. The AI reads the description field and decides when to activate it automatically — no hooks, no configuration, no ceremony.

---
name: tracking-tasks
description: Enforces disciplined task tracking. Use when starting any coding task...
---

# Tracking Tasks

## Overview
Every action must be tracked...

Drop it in the skills directory. It just works.

For the full spec, see Extend Claude with skills.

Skill-Check Hook

The plugin includes a UserPromptSubmit hook that fires on every user message, reminding the agent to check skills before acting. This is a mechanical safety net — skills exist but agents rationalize past them. The hook makes skipping harder.

Plugin users get this automatically. For manual installs, add to ~/.claude/settings.json:

{
  "hooks": {
    "UserPromptSubmit": [
      {
        "hooks": [
          {
            "type": "command",
            "command": "echo 'SKILL CHECK: Before acting, scan the skills list. Invoke tracking-tasks + any matching domain skills BEFORE exploring, reading, or writing anything.'",
            "timeout": 5
          }
        ]
      }
    ]
  }
}

Contributing

Want to add a skill? See AGENTS.md for conventions:

• Gerund naming — tracking-tasks, not task-tracker
• Test with a subagent before merging — if it doesn't trigger, it doesn't ship
• Update README + AGENTS.md with every new skill
• Conventional commits with gitmoji — see the committing-code skill

License

MIT