@daanvandenbergh/claudekit

This is not an SDK and not a wrapper around the Claude / Anthropic API. It ships no runtime code and makes no API calls. It's a content-only collection of reusable Claude Code skills and CLAUDE.md rules that Claude Code agents load - you drop them into any project.

Reusable Claude Code skills and CLAUDE.md rule snippets, shipped as plain content. You install the package as a dev-dependency, then symlink the skills you want into your project's .claude/skills/ and @-import the rules you want into your CLAUDE.md. Because both are references into node_modules, they auto-update whenever you bump the package - no re-copying.

What's inside

skills/
  ts/      TypeScript / general code-quality skills
    audit             phased, production-grade single-module audit (security, quality, logic, tests) that fixes what it finds
    audit-deep-logic  whole-system audit of the bugs that live in the SEAMS between modules
    audit-security    project-agnostic WHOLE-PROJECT security audit: real scanners (deps/secrets/SAST/IaC) + its own LLM whole-repo review + a threat model + the project's own invariants, adversarially triaged; report-only, --fix opts in
    audit-tests       hunt untested logic and write edge-case-exhaustive unit tests to total coverage
  legal/   legal / compliance skills
    policy-writer     create or verify-and-update code-grounded privacy/cookie/terms/DPA/imprint/AI-disclosure policies
  memory/  agent-memory skills
    memory-store      persist agent memory (facts, preferences, decisions, project state) into a de-duplicated, agent-readable Markdown store (--store) and keep it organized (--optimize)
  web/     web / SEO skills
    seo               audit the whole site for SEO and auto-fix what's safe (titles, meta, structured data, canonical/robots/sitemap, Core Web Vitals, answer-engine optimization); the de-index-class of changes is always confirmed first, --scan reports only
rules/     CLAUDE.md snippets you compose into a project's CLAUDE.md
  core_principles.md  simplicity-first / no-laziness / minimal-impact
  workflow.md         plan-mode, subagents, self-improvement loop, verification, task management
  todo.md             the deferred-work (TODO.md) convention

Install

npm install --save-dev @daanvandenbergh/claudekit

Link a skill (macOS / Linux symlink, auto-updating)

Claude Code discovers skills under .claude/skills/<name>/SKILL.md. Symlink the skill directory out of node_modules so an npm update of this package updates the skill in place - exactly how @daanvandenbergh/scribekit is linked.

Run from your project root (the symlink is relative, so it survives moving the repo):

mkdir -p .claude/skills

# a TypeScript skill
ln -s ../../node_modules/@daanvandenbergh/claudekit/skills/ts/audit \
      .claude/skills/audit

ln -s ../../node_modules/@daanvandenbergh/claudekit/skills/ts/audit-deep-logic \
      .claude/skills/audit-deep-logic

ln -s ../../node_modules/@daanvandenbergh/claudekit/skills/ts/audit-security \
      .claude/skills/audit-security

ln -s ../../node_modules/@daanvandenbergh/claudekit/skills/ts/audit-tests \
      .claude/skills/audit-tests

# a legal skill
ln -s ../../node_modules/@daanvandenbergh/claudekit/skills/legal/policy-writer \
      .claude/skills/policy-writer

# a memory-store skill
ln -s ../../node_modules/@daanvandenbergh/claudekit/skills/memory/memory-store \
      .claude/skills/memory-store

# a web / SEO skill
ln -s ../../node_modules/@daanvandenbergh/claudekit/skills/web/seo \
      .claude/skills/seo

The ../../ is because .claude/skills/<name> is two directories deep, so it climbs back to the project root before descending into node_modules. Verify with ls -l .claude/skills (each entry should point into node_modules/...) and restart Claude Code so it re-scans.

Include a rule in your CLAUDE.md

Claude Code inlines any @path reference in CLAUDE.md at load time. Point it straight at the file in node_modules so the rule text auto-updates with the package:

# CLAUDE.md

@node_modules/@daanvandenbergh/claudekit/rules/core_principles.md
@node_modules/@daanvandenbergh/claudekit/rules/workflow.md
@node_modules/@daanvandenbergh/claudekit/rules/todo.md

Add only the rule files you want; each is self-contained. The path is relative to the CLAUDE.md file, so keep CLAUDE.md at the project root (next to node_modules).

Some skills also ship a claude-rules.md meant to be imported the same way, so the rule rides in context every session (not only when the skill runs). If you use /audit-security, import its rule so maintaining claude/audit-security/rules.md (the project's own security invariants) stays a standing obligation:

@node_modules/@daanvandenbergh/claudekit/skills/ts/audit-security/claude-rules.md

Updating

npm update @daanvandenbergh/claudekit

Symlinked skills and @-imported rules pick up the new content automatically - nothing to re-link.