ClaudeKit Skills - Setup Guide¶

Source: mrgoonie/claudekit-skills via skills.sh - 18.0K combined installs across 33 indexed listings; first seen 2026-10-31 (evening sweep)

GitHub: mrgoonie/claudekit-skills - 2,230 stars, no LICENSE file, branch main, last pushed 2026-04-03; 45 SKILL.md files under .claude/skills/, 31 of them at the top level and 14 nested inside three grouped packs (debugging/ with four sub-skills, problem-solving/ with six, document-skills/ with docx, pdf, pptx and xlsx), plus a shared common/ directory and 12 plugin manifests under plugins/

Category: Full-stack delivery toolkit for agents: document skills, UI styling, MCP building and code review

Quality Tier: 🟡 Yellow - a Gen Agent Trust Hub Warn on debugging, a Snyk Fail on pdf, and no LICENSE file in the repository

ClaudeKit is one person's working notes on Claude Code turned into a skills collection, and the README says as much: six months of digging into the tool, published as a free tier with a paid bundle at ClaudeKit.cc behind it. What ships here is broad rather than deep in any one place. Backend, frontend, databases, DevOps, testing, document processing, media processing, 3D, Shopify, payments, MCP building and a set of Anthropic-authored meta skills all sit in the same tree, and the marketplace groups them into 12 installable plugin categories.

The structure is a .claude/skills/ directory with 45 SKILL.md files, three of the groups being containers rather than single skills: debugging pairs a router skill with root cause tracing, defense in depth, systematic debugging and verification before completion; problem-solving pairs a router with six named techniques plus a when-stuck dispatcher; document-skills holds docx, pdf, pptx and xlsx as four separate skills. That fan-out matters when you install, because a copy of a parent directory has to travel with its children or the router skills break.

Installation¶

The README's recommended route is the Claude Code plugin marketplace, which installs a named category rather than the whole tree:

/plugin marketplace add mrgoonie/claudekit-skills
/plugin install ai-ml-tools@claudekit-skills
/plugin install web-dev-tools@claudekit-skills
/plugin install devops-tools@claudekit-skills

The available categories are ai-ml-tools, web-dev-tools, devops-tools, backend-tools, document-processing, debugging-tools, problem-solving-tools, platform-tools, meta-tools, media-tools, research-tools and specialized-tools, each backed by a directory under plugins/ and described in .claude-plugin/marketplace.json. The plugin route is Claude Code only, and the repository documents no skills CLI command, so npx skills add is not an option here and anything that claims otherwise for this repo is guesswork.

The README's manual route is marked deprecated but still works, and it is the one Hermes needs, adjusted to copy the skills tree rather than the whole .claude directory:

git clone --depth 1 https://github.com/mrgoonie/claudekit-skills.git /tmp/claudekit-skills
cp -r /tmp/claudekit-skills/.claude/skills/. ~/.hermes/profiles/corpusiq/skills/

INSTALLATION.md is worth reading before the first run, because most skills assume tools that are not in the repo. ai-multimodal is the only skill with Python packages, listed in scripts/requirements.txt (google-genai, pypdf, python-docx, markdown, Pillow, python-dotenv). media-processing needs FFmpeg and ImageMagick, devops wants Wrangler, Docker and the Google Cloud CLI, better-auth and shopify want Node.js 18 or newer with the Better Auth and Shopify CLIs, and databases wants a PostgreSQL client and the MongoDB shell.

What It Provides¶

Skill	Installs	Use For
backend-development	2,307	APIs, auth, data access, CI/CD and microservice patterns in Node.js, Python, Go or Rust, with OWASP Top 10 coverage and scalability guidance.
sequential-thinking	1,798	Step-by-step reasoning that can revise and branch for multi-stage analysis, design planning and problem decomposition.
threejs	1,255	3D web work in Three.js on WebGL or WebGPU: scenes, shaders, PBR materials, VR and XR experiences, and product configurators.
devops	990	Deploys to Cloudflare Workers, R2 and D1, Docker, Cloud Run and GKE, Kubernetes with kubectl and Helm, plus CI/CD and security audits.
chrome-devtools	740	Browser automation and performance work through Puppeteer CLI scripts: screenshots, network traces, form automation and JavaScript debugging.
aesthetic	718	Visual design principles for interfaces, including inspiration capture through `chrome-devtools`, iterative design images through `ai-multimodal` and design documentation.
ui-styling	639	shadcn/ui and Tailwind component work: design systems, responsive layouts, accessible dialogs and tables, theming and dark mode.
problem-solving	576	A router across creative techniques (collision-zone thinking, inversion, pattern recognition, simplification cascades) with a `when-stuck` dispatcher for the right one.
debugging	573	Systematic debugging with four sub-skills: root cause tracing, defense in depth, systematic debugging and verification before completion.
media-processing	562	FFmpeg and ImageMagick pipelines: format conversion, codec choices, HLS and DASH manifests, thumbnails, filters and batch image work.
ai-multimodal	535	Google Gemini API work: audio transcription with timestamps, image understanding and generation, video analysis and PDF or document extraction.
code-review	518	Receiving review feedback with technical rigour, requesting reviews from a subagent, and gating completion claims on evidence.
context-engineering	499	Designing agent context: degradation patterns, compaction, masking and caching, memory architectures, multi-agent patterns and LLM-as-judge evaluation.
shopify	452	Shopify apps, extensions and themes: GraphQL and REST admin APIs, the Shopify CLI, Polaris components, Liquid, webhooks and billing.
web-testing	444	Playwright, Vitest and k6 across E2E, unit, integration, load, visual and accessibility testing, including flake and Core Web Vitals work.
databases	438	MongoDB and PostgreSQL schema design, queries, aggregation, indexing, migrations, replication, backup and administration.
frontend-design	419	Distinctive, production-grade frontend code that avoids the generic look of generated interfaces.
docs-seeker	405	Finding current technical documentation through the llms.txt standard, Repomix packing and parallel source exploration.
repomix	404	Packing repositories into single AI-readable files with include and exclude patterns, XML or Markdown output, and token counts.
better-auth	392	TypeScript authentication: email and password, OAuth providers, TOTP and SMS second factors, passkeys, sessions, RBAC and rate limiting.

The remaining 13 indexed listings range from 83 to 375 installs. The document skills sit at the bottom of that tail (docx 88, pptx 85, xlsx 84, pdf 83), which is worth knowing because they are the ones with the heaviest script bundles.

Why This Matters for Hermes Agents¶

The pitch for this pack is breadth with no setup cost. A single clone gives a profile the design, backend, DevOps, testing and document workflows that most delivery tasks touch, and the skills are written as workflows with guardrails rather than as documentation summaries, so the agent gets an instruction order instead of a knowledge dump. For a profile that does research, publishing and site work, the useful overlap is docs-seeker, repomix, context-engineering and code-review, which are the four that change how an agent reasons rather than what it can build.

code-review is the sharpest of those. It is explicitly about refusing performative agreement, requesting review through a subagent, and demanding evidence before any completion claim, which is the same failure mode most agent runs exhibit when they report a task done without running it. context-engineering covers the other half of the same problem: compaction, masking, caching and memory architecture, written by someone who was fighting context limits in Claude Code rather than describing them abstractly.

The meta skills are the reason to read before copying. skill-creator and template-skill show the author's own conventions for skill structure, which is useful as a comparison point for a house style, and docs-seeker implements the llms.txt lookup pattern that keeps a profile from asserting facts about a library it has not read.

Usage¶

You say	What happens
"Build the API for this feature"	`backend-development` picks the framework path, lays out routes, auth and data access, and adds the security and testing checklist.
"These two tests flake in CI"	`debugging` routes to root cause tracing or systematic debugging, then `verification-before-completion` demands a rerun as evidence before the fix is called done.
"Package this repo so you can reason about it"	`repomix` packs the tree with include and exclude patterns and reports a token count before the agent reads it.
"Find the current docs for this library"	`docs-seeker` tries the llms.txt standard first, falls back to parallel source exploration, and packs a GitHub repo when no documentation exists.
"Review this PR before I merge it"	`code-review` runs the three practices: technical review of the diff, a subagent review pass, and a verification gate that blocks a success claim without evidence.
"Make this dashboard look less generated"	`aesthetic` and `ui-styling` work the hierarchy, spacing and component layer, with `frontend-design` writing the code.
"Turn this PDF into structured data"	The `pdf` sub-skill under `document-skills` reads the form and table tooling in its `scripts/` folder; read that skill before running it, given its verdicts.

Verification¶

The raw path for the top skill returned HTTP 200 on branch main when this guide was written on 2026-10-31:

curl -sL -o /dev/null -w "%{http_code} backend-development\n" \
  https://raw.githubusercontent.com/mrgoonie/claudekit-skills/main/.claude/skills/backend-development/SKILL.md

Expect 200. Inspect the clone before copying, since the grouped packs are the part that silently breaks:

find /tmp/claudekit-skills/.claude/skills -name SKILL.md | wc -l
head -6 /tmp/claudekit-skills/.claude/skills/backend-development/SKILL.md
ls /tmp/claudekit-skills/.claude/skills/debugging

The count should print 45, the header should show name: backend-development with a license: MIT field in the frontmatter, and the listing should show four sub-skill directories beside SKILL.md. After the copy, ls ~/.hermes/profiles/corpusiq/skills/debugging should show the same four.

Security¶

skills.sh verdicts for sampled skills (verified 2026-10-31) - re-check the security pages on skills.sh before production use:

Skill	Gen Agent Trust Hub	Socket	Snyk
backend-development	Pass	Pass	Pass
debugging	Warn	Pass	Pass
frontend-design	Pass	Pass	Pass
skill-creator	Pass	Pass	Warn
pdf	Pass	Pass	Fail

The licensing position needs stating plainly, because it is the weakest part of this pack. There is no LICENSE file anywhere in the repository, which under default copyright means no grant to redistribute, while each SKILL.md carries license: MIT in its own frontmatter and .claude-plugin/marketplace.json declares "license": "MIT" for the ai-ml-tools plugin only. Treat the pack as MIT by intent, not by instrument: fine to read and use locally, unsettled if you plan to redistribute it, and worth an issue to the owner before you publish a derivative.

The pdf Fail is the one to respect. That sub-skill carries a script bundle that writes to real documents: filling form fields, annotating forms, converting PDFs to images, extracting field info and checking bounding boxes. None of that needs a network connection to cause damage if it is handed the wrong file, and the scanners are right to look harder at it than at a markdown-only skill. Read document-skills/pdf/scripts before letting an agent run it unattended.

The rest of the surface is ordinary development tooling. The 78 Python scripts concentrate in document-skills (32), then media-processing, databases and ai-multimodal at six each, with the remainder spread across web-frameworks, ui-styling, skill-creator and devops. The 22 JavaScript files are mostly chrome-devtools Puppeteer scripts (15), plus payment-integration (4) and a built mcp-management dist (3). The 78 XSD schemas belong to the OOXML validation sets under document-skills/docx and document-skills/pptx, and the 54 font files sit in ui-styling/canvas-fonts. The debugging Warn on Gen Agent Trust Hub and the skill-creator Snyk Warn are prompt-level flags rather than code findings; both skills are instructions about how to critique and how to write skills, and neither ships executables.

Integration with CorpusIQ¶

CorpusIQ is the connector layer that lets AI assistants read the systems a business already runs on - accounting, payments, CRM, email, files and analytics - and answer with the source named. It is read-only by design: it retrieves data and never writes to, or acts in, a connected system, which is what makes it safe to point at live accounts.

This page is part of the CorpusIQ directory, the reference we maintain for the MCP server and agent-skill ecosystem. Run alongside CorpusIQ, a third-party server like this one can be combined with a company's own business data in a single answer.

See what CorpusIQ connects

Limitations¶

Snapshot data, verified 2026-10-31: 17,978 combined installs across 33 indexed listings; 2,230 stars; no LICENSE file; last pushed 2026-04-03. Counts drift over time.
No LICENSE file in the repository. Per-skill frontmatter and one marketplace entry claim MIT, so the intent is clear, but the grant itself is missing.
The index and the tree disagree on one name: frontend-dev-guidelines carries 375 installs and has no directory in the current tree, while the tree's frontend-development/ directory has no indexed listing. Treat the pair as the same intent and read the directory.
Last pushed 2026-04-03, which is about seven months before this snapshot. Framework-specific material (frontend-development, web-frameworks, shopify) can lag the versions an agent is asked about, so verify against current docs before trusting version details.
The tree holds 45 SKILL.md files against 33 indexed listings, and the difference is structural: the debugging and problem-solving packs contribute ten sub-skill files that are counted once at parent level, claude-code, google-adk-python and frontend-development are in the tree but not in the index, and frontend-dev-guidelines is in the index without a directory.
Several skills need tooling that is not in the repo: FFmpeg and ImageMagick for media-processing, Wrangler, Docker and the Google Cloud CLI for devops, Node.js 18 or newer for shopify and better-auth, and Python packages plus a Gemini API key for ai-multimodal.

Skills Catalog | Skills Marketplace

Pro Tips¶

Install by group, not by file. debugging and problem-solving are routers whose value is in the sub-skill directories beside them, and a copy that takes only the parent SKILL.md leaves the agent with an index and no content.
Put code-review in the profile before anything else. It is the skill that makes an agent show a rerun instead of claiming success, and it changes the quality of every other task more than any build skill here.
Give context-engineering a read even if you never install it. Its compaction and masking sections describe the failure modes a long-running profile hits, and the vocabulary alone makes context problems easier to debug.
Read .claude-plugin/marketplace.json before using the plugin route. Category names and contents are defined there, and the plugin route installs a category rather than the tree, so what you get depends on which one you name.
Treat the pdf skill as code review material. Its scripts touch real documents, and its Snyk Fail is the only hard failure in the pack, so read the scripts before an unattended run.

← Skills Catalog | Skills Marketplace → Powered by CorpusIQ