Your AI coding agent restarts from zero every session. Agent Skills gives you a browsable catalog of standalone, versioned workflows — including this site from docs/. Install only what you need with one command. Your agent follows the same process every time.
You rewrite instructions for code review, releases, and planning on every project. Different chats produce different quality.
You retype the same checklist. Wording shifts between sessions and the agent's output quality drops.
Without a written workflow on disk, the agent cannot apply the same steps consistently across projects.
Each AI tool stores skills in different locations. You maintain duplicate instructions by hand.
Vague or varying instructions ship inconsistent patterns into code and team habits.
Without a fixed checklist, security and edge cases get skipped when time is short.
New contributors and new projects force you to rebuild prompts from memory or stale chat logs.
Pasting long instructions each session burns context and still omits the templates and guardrails that matter.
Agent Skills is a catalog of standalone workflows. Pick the capabilities you need. Install with one command. Your agent executes the documented process on every project.
Install only code-review, or only release-manager. Each skill ships as its own directory with SKILL.md, references, and scripts.
npx skills add or asm install. The command places files in the correct path for Claude Code, Cursor, Windsurf, Copilot, Codex, OpenCode, or Google Antigravity.
Skills include acceptance criteria, guardrails, templates, and references so the agent has the full context on disk.
Why it beats ad-hoc prompts: version-controlled, shareable with your team, and structured so agents load only the references the task needs.
Choose from categories: code quality, shipping, product planning, design, documentation, tooling.
Run the npx or asm command. Choose global install or project-local scope.
Your agent loads the skill by name and follows the exact workflow, references, and rules.
npx skills add https://github.com/luongnv89/skills --skill code-review
Install several at once: add --skill security-setup --skill code-review or --skill website-cloner. Latest tagged release in the changelog: v2.0.0.
Major catalog release — breaking rename, new shipping and modernization skills, orchestrator succession, and a measurable evals harness. This GitHub Pages home is the public catalog.
docs-generator is now doc-manager. Update install flags and bookmarks. Overlapping skills were consolidated into fewer entry points.
codebase-modernizer, issue-work-loop, diagram-generator, fork-upstream-sync, and herdr-agent-comms.
Orchestrator context succession plus an evals delta harness so catalog changes stay measurable.
Open source and designed to work in the tools you already use.
Code review, issue-work-loop, codebase-modernizer, release-manager, website-cloner suite, Herdr/tmux agent comms — each installs on its own.
Installers handle paths for Claude Code, Cursor, Windsurf, GitHub Copilot, OpenAI Codex, OpenCode, and Google Antigravity.
Free to use, inspect, fork, and contribute. No strings attached.
Catalog v2.0.0 on GitHub Pages — 42 installable workflows, suite-aware installers, HANDOFF succession, and evals harness.
Seven categories — plus multi-phase suites like website-cloner and diagram-generator. Install one skill or ten. They stand alone.
Structured reviews by severity using Code Smells and Pragmatic Programmer patterns.
Read-only whole-repo audit and a phased, testable modernization plan.
Version bump, changelog, tags, GitHub release, and package publish in one workflow.
Generate or reconcile Markdown docs to match the code, with citations and runbook checks.
Implementer and reviewer loop on one GitHub issue until the PR review is CLEAN.
Manage Herdr agent fleets: tile panes, message/wait/read, and steer any pane.
Six-phase orchestrator to analyze, plan, and rebuild an improved site.
No. Each skill is independent. Install only what your current project needs.
Any agent tool that supports skills. The installer handles file paths for Claude Code, Cursor, Windsurf, Copilot, Codex, OpenCode, and Google Antigravity automatically.
A skill is a structured workflow with references, templates, and quality checks — version-controlled and shareable. A prompt is a one-off instruction that lives in chat history.
No. Skills guide your agent during development. Nothing to deploy, no runtime dependencies in your app.
Yes. Follow the contributing guide and use the skill-creator scripts documented there. Same format, same validation tooling.
Yes. MIT licensed. Install, fork, and contribute at no cost.
Ship v2.0.0 — no bundle, no framework. One command installs a standalone workflow. Breaking: use --skill doc-manager, not docs-generator.
MIT licensed · open source · global or per-project scope