Skip to content

Latest commit

 

History

26 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

agents-skills

A collection of Agent Skills for Claude Code and Codex. Each skill is a SKILL.md file, sometimes with reference docs or scripts beside it, that the agent loads when your request matches its description.

Skills

Skill What it does Credit
cognitive-map Build a goal-oriented map of a topic's core concepts, dependencies, branches, and learning paths.
dev-guide Create and maintain a repository's developer or architecture guide, grounded in real code.
excalidraw Generate architecture diagrams as editable .excalidraw files, with optional PNG/SVG export. Adapted from ooiyeefei/ccc
git-commit Write Conventional Commits and split a working tree into clean, independent commits. Based on Conventional Commits
product-designer Turn a rough product idea into a testable, buildable MVP brief. HCD principles from The Design of Everyday Things
readme-writer Write or improve a project's README without fabricating commands, badges, or behavior. Based on makeareadme.com
teach Turn the current directory into a stateful, multi-session learning workspace with short, citation-grounded lessons. By Matt Pocock, mattpocock/skills (MIT)
unslop Cut AI tells from any writing and give it a human voice. Adapted from cursor/plugins (MIT)

cognitive-map

Build a cognitive map that helps a learner get oriented in an unfamiliar subject.

  • Identifies the central question and the fewest concepts you need before the rest makes sense.
  • Shows prerequisites, relationships, major branches, and alternative learning paths.
  • Covers the misconceptions beginners hit, what experienced people know not to do, and what to learn next.

Invoke with $cognitive-map in Codex, /agents-skills:cognitive-map when installed as a Claude Code plugin, /cognitive-map as a standalone Claude skill, or ask to "map this topic for me."


dev-guide

Create and maintain a repository's developer or architecture guide.

  • Grounds every description in code it has read.
  • Chooses the correct diagram type for each idea.
  • Links components to their source and traces real data and control flows.
  • Includes references for guide structure and diagrams.

Invoke with $dev-guide in Codex, /dev-guide in Claude Code, or ask to "update the dev guide" or "document this module."


excalidraw

Generate architecture diagrams as editable .excalidraw files by reading the codebase.

  • Maps components, services, databases, APIs, and data flows.
  • Uses cloud-provider color palettes for AWS, Azure, GCP, and Kubernetes.
  • Handles arrow routing, bindings, and staggering to reduce overlap.
  • Can export to PNG or SVG with Playwright.

Invoke with $excalidraw in Codex, /excalidraw in Claude Code, or ask to "create an architecture diagram."


git-commit

Write Conventional Commits and structure a working tree into clean commits.

Real git log from this repository after using the git-commit skill

The skill wrote every commit in this capture.

  • Reads the diff before composing a message.
  • Splits mixed changes into independent, self-contained commits.
  • Applies Conventional Commits types, scopes, breaking changes, and SemVer mapping.

Invoke with $git-commit in Codex, /git-commit in Claude Code, or ask to "commit this" or "write a commit message."


product-designer

Turn a rough product idea into a focused MVP brief.

  • Produces personas, jobs-to-be-done, requirements, user stories, acceptance criteria, a data model, metrics, risks, and milestones.
  • Makes requirements testable and traceable to user needs.
  • Applies human-centered design principles.

Invoke with $product-designer in Codex, /product-designer in Claude Code, or ask to "write a PRD" or "turn this idea into an MVP plan."


readme-writer

Write, overhaul, or fix a project's README.

  • Orders sections by how much the reader has committed so far: identity, proof, quickstart, depth.
  • Verifies commands, examples, configuration, and badges against the repository.
  • Adapts its structure for libraries, CLIs, research/ML projects, and web services.

Invoke with $readme-writer in Codex, /readme-writer in Claude Code, or ask to "make a README" or "document this project."


unslop

Edit prose so it stops reading like a language model wrote it.

  • Lists 31 tells in seven groups, each paired with the rewrite that fixes it.
  • Bans em dashes, colon-as-connector, restated bold labels, and title case headings.
  • Pushes for active voice, plain words, and sentences that state a fact rather than name a feeling.
  • Treats voice as part of the job. Stripping the patterns and stopping there leaves sterile writing, which is its own tell.

Invoke with $unslop in Codex, /unslop in Claude Code, or ask to "unslop this" or "make this sound human." It also works as a last pass over anything an agent just wrote.

Install for Codex

The repository doubles as a Codex plugin, so one marketplace install gets you all eight skills:

codex plugin marketplace add Punpun1643/agents-skills
codex plugin add agents-skills@Punpun1643

Start a new Codex session after installing. Run /skills to browse what landed, name one with $skill-name, or describe the task and let Codex pick.

Upgrade the Codex marketplace

Refresh the installed Git marketplace snapshot with its configured marketplace name:

codex plugin marketplace upgrade Punpun1643

The upgrade command expects Punpun1643, not the GitHub repository slug Punpun1643/agents-skills.

Install one standalone skill

Clone the repository, then symlink one skill into Codex's personal skills directory:

git clone https://github.com/Punpun1643/agents-skills.git
cd agents-skills
mkdir -p ~/.agents/skills
ln -s "$PWD/plugins/agents-skills/skills/readme-writer" ~/.agents/skills/readme-writer

Replace readme-writer with any skill name from the table above. Codex follows symlinked skill folders, so pulling the clone updates the installed skill.

To pin a fixed version instead, copy the folder:

cp -R "$PWD/plugins/agents-skills/skills/readme-writer" ~/.agents/skills/readme-writer

Install for Claude Code

Install the whole collection as a Claude Code plugin from its marketplace:

claude plugin marketplace add Punpun1643/agents-skills
claude plugin install agents-skills@Punpun1643

Claude Code namespaces plugin skills. Invoke the README skill with /agents-skills:readme-writer, or describe the task and let Claude pick.

Install one standalone skill

Clone the repository and create Claude Code's personal skills directory:

git clone https://github.com/Punpun1643/agents-skills.git
cd agents-skills
mkdir -p ~/.claude/skills

Install all skills as symlinks so they track the clone:

for d in cognitive-map cognitive-map-open dev-guide excalidraw git-commit product-designer readme-writer unslop; do
  ln -s "$PWD/plugins/agents-skills/skills/$d" ~/.claude/skills/"$d"
done

Or install a fixed snapshot by replacing ln -s with cp -R:

for d in cognitive-map cognitive-map-open dev-guide excalidraw git-commit product-designer readme-writer unslop; do
  cp -R "$PWD/plugins/agents-skills/skills/$d" ~/.claude/skills/"$d"
done

License

Released under the MIT License. The adapted Excalidraw and Unslop skills also include their upstream attribution: excalidraw, unslop.

Releases

Packages

Contributors

Languages