Agent Skills Collection

This repository is a growing set of skills based off the Agent Skills specifications. Each skill is a small, focused capability that can be plugged into AI agents to give them repeatable workflows, domain expertise, or tooling integrations.

Structure

• Skill directories: Each top-level folder (for example, product-management/, design/, etc.) groups related skills by domain.
• SKILL.md files: Every skill is defined in a markdown file named SKILL.md that:
  • Describes what the skill does and when to use it.
  • Defines the instructions the agent should follow.
  • Follows the Agent Skills specification so tools can discover and load it.

Installation & Usage

OpenAI Codex

Codex loads skills from .agents/skills (repo scope) and ~/.agents/skills (user scope), as described in the Codex skills docs (link).

You have two main options to install a single skill from this repo:

• Using the built-in installer (recommended when available in your Codex environment), e.g. for dsat-initialize-ds:

  $skill-installer install https://github.com/eshraw/agents-skills/tree/main/design/design-system-as-text/dsat-initialize-ds

• Manual copy into a skills directory, e.g. for assumption-audit in the current repository:

  git clone https://github.com/eshane-rawat/agents-skills.git /tmp/agents-skills
  mkdir -p .agents/skills/assumption-audit
  cp -R /tmp/agents-skills/product-management/assumption-audit/* .agents/skills/assumption-audit/
  rm -rf /tmp/agents-skills

  Or as a user-wide skill:

  git clone https://github.com/eshane-rawat/agents-skills.git /tmp/agents-skills
  mkdir -p ~/.agents/skills/assumption-audit
  cp -R /tmp/agents-skills/product-management/assumption-audit/* ~/.agents/skills/assumption-audit/
  rm -rf /tmp/agents-skills

In all cases, replace the example paths and destination folder names with the specific skill you want (for example, design/design-system-as-text/dsat-initialize-ds). Restart Codex if it doesn’t pick up the new skill automatically.

Claude Code (plugin marketplace)

Skills from this repo are available as installable Claude Code plugins via eshraw/skills-mkp, a marketplace that is automatically synced from this repo.

Add the marketplace and install any skill in two commands:

claude plugin marketplace add eshraw/skills-mkp
claude plugin install assumption-audit@skills-mkp

Claude (manual)

Clone the repo, then copy just the skill you want into Claude’s skills directory, e.g. for assumption-audit:

git clone https://github.com/eshane-rawat/agents-skills.git
cd agents-skills
mkdir -p ~/.claude/skills/assumption-audit
cp -R product-management/assumption-audit/* ~/.claude/skills/assumption-audit/

Point your Claude client or orchestration layer at ~/.claude/skills/assumption-audit as a single registered skill. Repeat with different target names/folders for other skills.

Cursor

To install these skills in Cursor using a remote GitHub rule:

• Open Cursor Settings → Rules

• In Project Rules, click Add Rule

• Choose Remote Rule (GitHub)

• Enter the GitHub repository URL, for example:

  https://github.com/eshraw/agents-skills
  

Cursor will then pull the skills from this repository according to the rule configuration. You can add additional remote rules if you want to point to other skill repos or forks.

Using These Skills

• In an agent environment: Point your agent framework or tools at this repo (or a subset of folders) so it can load SKILL.md definitions according to the Agent Skills spec.
• Manually: You can also open any SKILL.md and follow the documented workflow as a lightweight playbook for that domain.

Adding a New Skill

• Create a folder for the domain if it does not exist yet (for example, engineering/, research/).
• Add SKILL.md inside that folder, following the existing examples:
  • Front‑matter with name and description.
  • Clear sections for overview, when to use, and step‑by‑step instructions.
• Keep it focused: A skill should solve one concrete task or workflow well, rather than many loosely related ones.

Contributing & Maintenance

• Consistent format: Match the structure and tone of existing skills so agents can apply them predictably.
• Incremental growth: Treat this as a living library—add new skills as needs arise, and refine existing ones as workflows improve.
• Compatibility: When updating skills, avoid breaking changes to names and core behaviors that downstream agents may depend on.