Skip to content

Repository files navigation

agentic-user-skills

A monorepo of user-level agentic skills — cross-agent tools that improve the user's experience across any repo and any coding agent (Claude Code, OpenCode, Cursor, Codex…), rather than skills tied to one project's stack.

Other repos (e.g. agentic-qa-boilerplate, agentic-dev-boilerplate) install these at the user (global) level via their installers, so the same skill is never duplicated per project. This repo is the single source of truth.

Install a skill (user-level)

These skills follow the skills convention. Install one globally with:

bunx skills add --global https://github.com/upex-galaxy/agentic-user-skills mkd

A user-level skill loads automatically in every project — no per-repo wiring.

Skills

Skill Command What it is
agentic-audit An evidence-based project auditor: reads a whole repo (git history, context artifacts, agent config, tracker mirror, tests, CI) and drives the live app, then scores six axes 0-5 and writes a self-contained HTML evaluation. Two modes: subject (dev | qa | pair) picks the rubric, lens (external | internal) picks the register and deliverable. Read-only on the target.
mkd bun cli/index.ts (no install) MKD (Make Decision) — a browser-based decision-deck CLI: the AI writes a spec of items (decision / question / report / answerable table) and the CLI renders a Catch-Up-style deck — one screen per item, options with written justifications, a ★ recommended badge, custom option, skip, quotes, live stats, localStorage persistence. Default flow is non-blocking copy-paste (the user pastes the Result JSON into the chat as the execution contract); --wait keeps a blocking same-turn handshake. Successor of wokitoki/toki.

Layout

agentic-user-skills/
  skills/
    <skill-name>/
      SKILL.md            # frontmatter + when/what/how + first-run setup
      references/         # deep docs loaded on demand
      cli/                # OPTIONAL bundled CLI — decoupled, zero external deps
  package.json            # gate scripts (lint / types)
  tsconfig.json eslint.config.js .gitignore skills.sh.json

A skill is just a directory. It may be pure instructions (SKILL.md + references/) or bundle a self-contained CLI. Each skill is independently extractable.

Conventions

  • Decoupled CLIs. Any bundled CLI imports runtime built-ins only (Bun / node: built-ins) — zero external npm deps, relative imports only. This is what lets a skill travel via bunx skills add and stay independently extractable.
  • Zero host footprint. A bundled tool writes its output (caches, results, artifacts) under the user's home (e.g. ~/.mkd/), never the consumer repo's cwd.
  • No install step. A bundled CLI runs straight from the skill directory (bun <skill-dir>/cli/index.ts) — no compiled binary, nothing added to PATH.

Toolchain

  • Bun runs the TypeScript directly.
  • Gates: bun run types:check (tsc --noEmit) and bun run lint:check (ESLint, @antfu/eslint-config).

License

MIT

About

Monorepo of user-level agentic skills — cross-agent tools (Claude Code, OpenCode, Cursor, Codex) installed globally via bunx skills add. First skill: wokitoki (toki), a human-in-the-loop feedback CLI.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages