Find your next open-source project

Search by name, purpose, or keyword — or paste a GitHub repository URL

humanlayer/skills

Developer Tools

A collection of installable Claude Code and Codex skills for research, implementation, visual PRs, and agent-loop design.

5,617
190
TypeScript
MIT
7 months old

What this is

This repository collects skills for coding agents, published by HumanLayer. A skill is an installable workflow that an agent can invoke from the command line, and these cover several development tasks: researching a change and turning it into a design document or implementation, explaining a topic with diagrams and HTML artifacts, generating a visual outline in a pull request, rewriting CLAUDE.md for better instruction adherence, narrowing React prop types, and building prompted agent loops or control loops.

The main audience is developers using Claude Code or Codex who want repeatable, structured workflows rather than one-off prompts.

Install and use a skill

The README's install pattern adds one skill at a time through the skills CLI:

npx skills add humanlayer/skills --skill SKILLNAME

After that, invoke the skill in your coding session. For example, to install and run the research-plan-implement workflow:

npx skills add humanlayer/skills --skill rpi

Then call /rpi with a task, or request a particular phase:

/rpi research how the collections connect to the database
/rpi make a PRD
/rpi what's next .agents/artifacts/my-change
/rpi implement .agents/artifacts/my-change

The RPI skill explicitly supports choosing a path: design discussion, PRD followed by TDD, technical design, or research followed by implementation. Each phase runs in a fresh session by default. Codex users call $rpi instead of /rpi.

Other skills follow the same two-step pattern. Install show-me, visual-pr, improve-claude-md, narrow-react-prop-types, build-iterated-agentic-loop, or design-control-loop with npx skills add humanlayer/skills --skill <name>, then invoke it with the matching slash command inside the project. design-control-loop is an interview-driven skill that produces locally runnable components plus a scheduled coding-agent workflow.

Constraints and caveats

  • The commands are npx-based, so Node.js and network access to the npm registry are required.
  • Codex's default sandbox may make .agents read-only. The RPI skill's documentation says to grant write access to .agents/artifacts or pass a writable task directory such as artifacts/my-change; the skill asks before writing to a blocked destination.
  • narrow-react-prop-types is React-specific. The other skills are not described as tied to a language in the provided README.
  • Skills are invoked differently in Codex ($name) than in Claude Code (/name).

HotRepo's take

The repository appeared on the all-language daily list in September 2026 with 1,141 stars gained on September 5, 451 on September 6, and 451 on September 7, moving from 2,563 to 3,360 stars over that window. On November 4 it recorded 30 daily stars and rank 13 on the TypeScript list, with 5,592 stars and 186 forks. The earlier burst and the later single-day TypeScript rank are different list scopes, so they don't establish a continuous trend, but they do show the project entered HotRepo's records through a sharp multi-day run before settling onto a language-specific list.

Project source and documentation

Written by AI from the project's README, updated . The project's own docs are the source of truth.