essential-skills

Homepage · npm

Interactive CLI that installs a curated set of agent skills into Cursor, Claude Code, Codex, GitHub Copilot, Gemini CLI, OpenCode, Windsurf, and other agents that read skills from .agents/skills.

Browse skills, compare the Quick and Full packs, and see how install works on the homepage.

Requires Node.js 20.12 or later.

Usage

npx essential-skills

The CLI walks you through:

1. Install or clear skills
2. Global (home directory) or project (current repo)
3. Which agents to target
4. Which skills to copy — via preset or manual selection

At any later step, press ← to go back. Esc cancels without changes.

Global installs land in each agent's home-directory skills folder. Project installs go into the matching folder in the current repo (for example .cursor/skills or .agents/skills).

On install, if a skill with the same name is already present, the CLI asks whether to override it or cancel. After a successful install it prints a short catalog of what was copied. Clear only removes skills from this pack — other skills in the same folders are left alone.

After installing, restart your agent session. Run /show-skill-catalog anytime to see the same list.

Presets

Pick a preset during install, or choose skills manually.

| Preset | Best for | Skills | | --- | --- | --- | | Quick | Pet projects, prototypes, fast iteration | 13 skills — planning and hygiene, no TDD or test overhead | | Full | Production apps where quality and tests matter | All 21 skills | | Manual | Mix and match | Pick each skill; high-overhead ones are labeled slow or moderate in the picker. Spec-driven context is one row and always copies /specify-context and /create-feature-spec together. |

Quick preset (13 skills)

grill-me, specify-context, create-feature-spec, show-skill-catalog, find-skills, fix-tech-debt, request-refactor-plan, apply-solid-principles, create-commit, fix-lint, apply-prettier, apply-style-guide, write-handoff

Keeps alignment before building (grill-me, spec-driven context) and a SOLID pass on demand, but skips TDD, test-writing, review, Storybook, and architecture skills so tasks resolve faster.

Full preset (21 skills)

Everything in Quick, plus:

review-code, develop-with-tdd, write-unit-tests, write-e2e-tests, write-storybook, apply-best-practices, feature-sliced-design, use-hybrid-folder-structure

Use Full when testing, review, and architecture guidance are non-negotiable.

Task overhead

Installing every skill makes the agent heavier on implementation work. These add the most time:

| Skill | Overhead | Why | | --- | --- | --- | | grill-me | Medium | Interview rounds and plan confirmation before coding | | review-code | High | Two parallel sub-agents over the full diff | | specify-context | Medium | Interview rounds to fill project context files, one file per invocation | | create-feature-spec | Medium | Interview rounds to write a unit spec and update the progress tracker | | develop-with-tdd | High | Failing test first; run the suite before and after each change | | write-unit-tests | High | Writes and runs unit tests | | write-e2e-tests | High | E2E specs are slow to author and run | | fix-tech-debt | Medium | Scan debt docs, domain inventory, selection, then grilling before fixes | | request-refactor-plan | Medium | Interview rounds, then a GitHub issue with a tiny-commit plan | | apply-solid-principles | Medium | Can split modules and invert dependencies across several files | | apply-best-practices | Medium | 70 React/Next.js rules; can trigger broad refactors | | feature-sliced-design | Medium | Architecture migrations and layer boundaries | | use-hybrid-folder-structure | Medium | Multi-file folder refactors |

Everything else is on-demand (commit, lint, format, handoff, catalog) and stays out of the way until invoked.

Included skills

| Skill | What it does | Preset | | --- | --- | --- | | grill-me | Interview before implementation; skip only pure Q&A or read-only review with no follow-up | Quick, Full | | review-code | Two-axis review of changes against repo standards and the spec | Full | | specify-context | Bootstrap docs context files, merge AGENTS.md/CLAUDE.md, grill one unfilled file per run | Quick, Full | | create-feature-spec | Grill a unit spec under feature-specs and update the progress tracker (installs with specify-context) | Quick, Full | | show-skill-catalog | List this pack and route you to the right skill | Quick, Full | | find-skills | Discover and install skills from the open ecosystem | Quick, Full | | fix-tech-debt | Discover README/TODO/FIX debt, group by domain, pick items, grill an implementation plan | Quick, Full | | request-refactor-plan | Interview, then file a GitHub issue with a tiny-commit refactor plan | Quick, Full | | apply-solid-principles | Apply SRP, OCP, LSP, ISP, and DIP to named or changed modules | Quick, Full | | create-commit | Split work into logical conventional commits; uses agent session history when available | Quick, Full | | fix-lint | Fix Biome or ESLint issues on named or changed files | Quick, Full | | apply-prettier | Format named or changed files with the project's Prettier | Quick, Full | | apply-style-guide | Apply a repo style guide, or Google's language guide if none | Quick, Full | | write-handoff | Compact the current conversation for another agent to continue | Quick, Full | | develop-with-tdd | Write the failing test first, then the implementation | Full | | write-unit-tests | Write unit tests for named units | Full | | write-e2e-tests | Write e2e tests for files you name | Full | | write-storybook | Write Storybook stories for named components | Full | | apply-best-practices | React and Next.js performance guidelines from Vercel Engineering | Full | | feature-sliced-design | Feature-Sliced Design (FSD) v2.1 from fsd.how | Full | | use-hybrid-folder-structure | Hybrid frontend layout: responsibility first, feature second | Full |

License

MIT