Search skills/

Claude Code Infrastructure Showcase Claude skills

A reference setup for Claude Code: hooks that auto-activate skills, a skill-rules.json config, and Node/Express/React stack guides to copy into your project.

diet103/claude-code-infrastructure-showcase 9 skills Planning, TDD & debuggingSkill & agent tooling
Our verdict Recommended with caveats

A well-organized reference for skill auto-activation, but the content is tied to one stack, the hooks need bash/WSL2, and it is meant to be adapted rather than installed as-is.

3/5 Depth expertise shipped ! Security Broad access 4mo Maintenance Updated 4 mo ago 10K Adoption stars OK License MIT, commercial OK

What Claude Code Infrastructure Showcase does

This is a copy-from reference library built from the author's months of work on a TypeScript microservices project. The main idea is a hook-based system that suggests or enforces the right skill when you type a prompt or edit a file, driven by a skill-rules.json file that maps keywords and file paths to skills. By default the matching is regex and keyword based; an optional mode asks Gemini, OpenAI or Anthropic to classify your intent instead.

The repo ships nine skill folders (the .claude and .agents copies are identical): backend-dev-guidelines for Express, Prisma and Zod layering, frontend-dev-guidelines for React 18, MUI v7 and TanStack, error-tracking for Sentry v8 patterns, skill-developer for writing your own skills and rules, and a verify-setup command. The two stack guides carry 10 or 11 reference files each, and skill-developer carries six. A setup.ts wizard copies the .claude directory into your project, detects your stack, installs hook dependencies and runs eight health checks. The stack guides are written for one specific stack, so they are only useful as-is if you use it; otherwise treat them as templates. The README says the hooks are bash scripts, so Windows users need WSL2. The README also mentions Codex support.

Good for

  • Developers who want to build skill auto-activation hooks for Claude Code
  • Teams on Node/Express/Prisma or React/MUI/TanStack stacks
  • Authors learning the 500-line rule and progressive-disclosure skill layout

Not for

  • People who want a ready-to-use skill pack for another stack
  • Plain Windows setups without WSL2

What's inside depth 3/5

9 skill folders (two duplicate copies of 4 skills plus a command) averaging about 10 KB, with 10-11 reference files in the main guides, but no scripts inside the skills, no tests and no evals in the skills.

Skills
9
Reference docs
54
Scripts
0
Tests / evals
0
Templates
0
Avg SKILL.md
9.5 KB

Counted from the repository file tree. How we score depth.

Dependencies and costs

  • Node.js 18+ and npm
  • bash (macOS, Linux or WSL2)
  • jq for optional Stop hooks
  • Optional Gemini, OpenAI or Anthropic API key for AI classification mode
  • Sentry, if you follow the error-tracking skill
  • Telemetry: No outbound telemetry found in the README. The README mentions a local activation log (state/metrics.jsonl). The optional AI mode sends your prompts to the provider whose key you configure; it is off by default.

Security review

Broad access

This skill asks for more access than most, or sends data to another service. That is usually a design choice rather than bad intent, but you should know before you install it.

  • Uploads your data by defaultThe PreToolUse guard sends the first 1000 characters of each edited code snippet to an LLM provider whenever GEMINI/OPENAI/ANTHROPIC_API_KEY (or OLLAMA_BASE_URL) is found in the environment, independent of skill_activation_mode (default 'disabled'); only SKIP_PRETOOLUSE_AI=true stops it. The hook wrapper sources ~/.bashrc and .claude/hooks/.env to find those keys..claude/hooks/skill-verification-guard.ts:319
  • Pre-approves broad shell commandsThe slash command frontmatter pre-approves Bash(xargs:*) and Bash(sed:*) alongside cat/awk/grep/sort, which can run arbitrary commands or edit files without a prompt..claude/commands/route-research-for-testing.md:4

A set of Claude Code skills plus hooks that run on every prompt, edit and session stop to suggest or force skills; by default it uses local regex matching and local state files. If you add a Gemini, OpenAI or Anthropic key (or have one in ~/.bashrc, which the hooks source), prompts and the first 1000 characters of edited code are sent to that provider, and the edit analysis ignores the 'disabled' AI setting. Setup installs npm packages and can overwrite your Neovim/vim config if you opt in.

Worth knowing

  • installs_hooksThe shipped settings.json registers hooks documented in the README on every user prompt, every Edit/MultiEdit/Write, every Skill call and every Stop; setup.ts copies it into the target project. session-doc-updater is installed by default but no-ops until session indexing is set up..claude/settings.json:2
  • Makes network requestsOptional AI classification and embeddings call Gemini, OpenAI (or OPENAI_BASE_URL), Anthropic or a local Ollama endpoint with the user's own key; prompts are sent when skill_activation_mode is changed from 'disabled'..claude/hooks/providers/gemini-provider.ts:25
  • Installs packages or toolsThe setup wizard runs npm install in .claude/hooks (tsx, better-sqlite3, minimatch, optional @google/genai, openai, @anthropic-ai/sdk).setup.ts:598
  • Changes files outside your projectIf the user opts in, setup.ts copies init.lua to ~/.config/nvim and vimrc to ~/.vimrc (overwriting existing files); README also suggests adding exports to ~/.bashrc.setup.ts:611
  • Relies on a CLI or package we could not reviewAn optional (not registered by default) Stop hook launches the claude CLI with the auto-error-resolver agent on changed services..claude/hooks/trigger-build-resolver.sh:70

What we checked

The repository (133 files) on Oct 29, 2026. Not a full review: we read 22 of 40 code files; it relies on code outside the repository: npm packages installed by setup (tsx, better-sqlite3, minimatch, @google/genai, openai, @anthropic-ai/sdk, typescript); package-lock.json pins protobufjs 7.5.4 which the scanner lists with known advisories; External LLM services (Gemini, OpenAI, Anthropic, Ollama) that receive prompts/code when AI mode or a provider key is present; Not read in full: setup.ts, skill-activation-prompt.ts, skill-verification-guard.ts, error-handling-reminder.ts, tsc-check.sh, trigger-build-resolver.sh, stop-build-check-enhanced.sh, lib/{metrics,session-parser,session-state,types,vector-store}.ts, providers/{ai-provider,parse-llm-json}.ts, .claude/scripts/{index-sessions.ts,verify-setup.sh}, editor-config/init.lua.

NVIDIA SkillSpector static scan: 237 pattern hits (MCP Rug Pull 76, Supply Chain 53, Agent Snooping 31, Tool Misuse 27, Privilege Escalation 19, Rogue Agent 19). Most HIGH hits are documentation: example DELETE routes and .env configuration in backend guideline docs, troubleshooting commands that read or delete the skill's own state under .claude/, and 'concealed executable artifact' flags on the repo's own hook files. The protobufjs CVE comes from the lockfile of the optional @google/genai dependency; no obfuscation or hidden exfiltration was found in the files read.

A review, not a certification. Skills change after we check them — read the files before installing.

License

MIT — commercial use allowed.

Can it run unattended?

Not really. It is an interactive prompt-time and edit-time guidance system, not a loop with stop conditions or resumable task state.

Alternatives to Claude Code Infrastructure Showcase

Superpowers

obra/superpowers

A 15-skill development methodology that makes coding agents clarify the goal, write a plan, work through it with subagents and verify before calling it done.

Recommended Check first Depth 3/5

A very large Claude Code harness: 311 skills, 71 agents, commands, hooks, memory, rules and a scanner, with adapters for Codex, Cursor and others.

With caveats Broad access Depth 3/5

Planning with Files

OthmanAdi/planning-with-files

Keeps task_plan.md, findings.md and progress.md on disk and re-injects them via hooks, so a plan survives /clear, compaction and crashes.

Recommended Broad access Depth 4/5

Agent Skills (Addy Osmani)

addyosmani/agent-skills

25 engineering-lifecycle skills and 9 slash commands (spec, plan, build, test, review, ship) that make coding agents follow senior-engineer workflows.

Recommended Check first Depth 3/5

FAQ

What is the Claude Code Infrastructure Showcase?

It is a reference repository by diet103 showing how to make Claude Code skills activate automatically using hooks and a skill-rules.json config. It also includes sample stack guides, agents and a dev-docs pattern. The README says it is not a working application; you copy parts into your own project.

How do I install it?

Clone the repo and run npx tsx setup.ts pointing at your project, or copy the .claude folder over and run npm install in .claude/hooks. The README requires Node.js 18+ and says the hooks need macOS, Linux or WSL2.

Do I need an API key?

No. The default mode uses regex and keyword matching with no API calls. An optional AI classification mode accepts a Gemini, OpenAI or Anthropic key and falls back to regex if the call fails.

Is it safe to install?

The skills are markdown guides. The risk is in the hooks and setup wizard, which run code on your machine on every prompt and edit. We could not inspect those files in our copy, so read them before enabling.

Sources: GitHub repository (stars, license, file tree, last push Jul 13, 2026). Review based on the README and SKILL.md files as published.