Skip to content

UmayerK/clanker-code

Repository files navigation

clanker-code

npm version CI license node

Claude Code in one command. Batteries, hooks, and guardrails included.

clanker-code transforms any repository into a fully-configured Claude Code workspace in under a minute. It installs curated skills, specialized agents, slash commands, safety hooks, and pre-wired local-first MCPs — all tuned for one thing: shipping code.


Quickstart

cd your-project
npx clanker-code init
claude

That's it. Your repo now has:

  • 4 pre-wired local-first MCPs (+ Serena auto-added if uv is detected) — no signup, no API keys
  • 26 slash commands — headlined by /vibe (fuzzy idea → working app, end to end), plus /brainstorm, /implement, /debug, /test, /feat, /select-tool, /pm, /recommend, /reflect, /revise-claude-md, /mcp-help, and more
  • 22 specialized agents — frontend-builder, backend-builder, debugger, reviewer, planner, tester, code-explorer, code-architect, security-engineer, architect, and 13 more
  • 46 curated skills — workflow, quality, security, git, framework-specific, plus framework patterns (persona auto-activation, wave orchestration, global flags, UC mode, semantic editing, TDD, API design review, dependency audits, CLAUDE.md maintenance)
  • Global command flags--strategy systematic|agile|enterprise, --depth shallow|normal|deep, --parallel, --validate, --uc (ultracompressed)
  • 8 safety + ergonomics hooks — auto-format, destructive-command guard, secret-leak guard, spec-awareness, git-context, SessionStart(compact) re-inject, and more (plus 5 opt-in)
  • Native statusline — branch, workspace, and model visible in your terminal
  • Stack-tailored CLAUDE.md — auto-detects Next.js, React, Node API, Python API, and more
  • specs/ scaffolding — spec-first workflow out of the box

The /vibe command

/vibe a team todo app with daily standup summaries

One command. Claude asks a few clarifying questions (Socratic, one at a time), writes the spec, plans it, builds it in parallel across specialized agents, tests with a real browser, reviews against OWASP and project standards, and stops with a suggested commit message. No auto-commits, no auto-pushes — you stay in control.


What problem does this solve?

Claude Code is powerful — and it ships as a blank slate. Every developer reinvents the same CLAUDE.md, the same skill bundle, the same hook configuration. Setup takes hours. Teams drift into inconsistent configs. MCPs require hand-editing JSON files.

clanker-code is the missing install layer. One command. Opinionated defaults. Zero dotfiles to copy-paste.

Why clanker-code

  • One command to ship-readynpx clanker-code init and you're done.
  • Curated, not maximal — every skill/agent/command/hook earns its slot. Startup token budget is deliberately kept under ~20K.
  • Safety-first — destructive-command guard, secret-leak guard, gitignore warnings, and SessionStart re-inject ship active by default.
  • Framework-aware — detects Next.js, React, Node APIs, Python APIs, Rust, Go; installs only the content relevant to your stack.
  • Interactive updatesclanker update does a 3-way merge so your customizations survive upgrades.
  • Cross-platform — native Windows, WSL, macOS, Linux. No Python or Bun dependencies.

Commands

# Install into the current repo (with interactive stack detection)
npx clanker-code init

# Common variants
npx clanker-code init --minimal         # CLAUDE.md + specs/ only
npx clanker-code init --setup-only      # specs/ scaffolding only
npx clanker-code init --no-mcps         # skip MCP configuration
npx clanker-code init --no-hooks        # skip hook configuration
npx clanker-code init --force           # overwrite without prompting
npx clanker-code init --dry-run         # preview what would be written

# Update to latest shipped version (interactive 3-way merge)
npx clanker-code update

# Discover and install more MCPs
npx clanker-code mcp-help                # interactive browse
npx clanker-code mcp-help list           # list all known MCPs
npx clanker-code mcp-help search <query> # search registry
npx clanker-code mcp-help show <name>    # details + setup instructions
npx clanker-code mcp-help add <name>     # guided install (prompts for keys)
npx clanker-code mcp-help remove <name>  # uninstall

# Health check
npx clanker-code doctor

# Meta
npx clanker-code help
npx clanker-code version

What gets installed

Default MCPs (4 + 1 conditional)

All local-first, no signup, no API keys.

MCP What it does
sequential-thinking Structured reasoning for architecture, debugging, complex problems
context7 Up-to-date library docs (kills hallucinated API calls)
memory Local knowledge graph — persist project facts across sessions
playwright Real browser automation for UI verification (downloads Chromium on first use)
serena (auto-added if uv is installed) LSP-backed semantic code editing — renames, find-usages, impact analysis

If uv is not installed, Serena is skipped with a hint for how to enable it. Install uv, then run npx clanker-code mcp-help add serena.

Need more? clanker mcp-help has a curated registry of 17+ MCPs across categories — databases, code intelligence, GitHub, Notion, Linear, Figma, and more — with guided setup for anything needing an API key.

Hooks

Active by default (8):

Hook What it does
Destructive Bash guard Blocks rm -rf /, force-push to main, DROP TABLE, fork bombs, mkfs, dd-to-device
Secret leak guard Blocks writing content with common API key patterns
Gitignore warning Warns before writing to gitignored paths (prevents lost work in .env, build/)
Auto-format Runs detected formatter (prettier, biome, ruff, black) on edited files
Spec-awareness Injects specs/ file list into context automatically
Git context Injects current branch + git status --short into every prompt
SessionStart(compact) re-inject Re-injects project context after auto-compaction
Stop notification Terminal bell + OS notification when a long task completes

Opt-in via CLANKER_HOOKS_EXTRA=on:

  • Session log on Stop (appends to .claude/sessions/YYYY-MM-DD.md)
  • Auto-typecheck after TS edits (background tsc --noEmit)
  • Package install reminder after package.json changes
  • PreCompact transcript backup
  • ExitPlanMode auto-approve

Kill switches:

  • CLANKER_HOOKS=off — disable all clanker hooks globally
  • CLANKER_HOOK_<NAME>=off — disable a specific hook (e.g., CLANKER_HOOK_AUTO_FORMAT=off)

Slash commands

26 commands. Every command supports the global flag set (--strategy, --depth, --parallel, --validate, --uc).

/vibe         /brainstorm   /implement    /analyze      /debug
/improve      /document     /test         /design       /build
/cleanup      /estimate     /explain      /git          /index
/load         /save         /reflect      /help         /research
/mcp-help     /feat         /select-tool  /pm           /recommend
/revise-claude-md

Power commands:

  • /vibe <idea>fuzzy idea → complete working app, end to end
  • /feat <name> — focused feature workflow with 3 parallel design approaches (minimal/clean/pragmatic)
  • /pm <goal> — multi-agent orchestration with task breakdown
  • /select-tool <task> — explicit MCP/tool selection based on task complexity
  • /recommend <goal> — fuzzy-ask to concrete-command recommendation
  • /reflect [--depth deep] [--validate] — semantic validation with confidence scoring
  • /index --brief — compress repo into a <3K-token brief for cheap downstream context
  • /revise-claude-md — audit and refresh CLAUDE.md against current codebase

Agents (22)

12 persona agents: architect, analyzer, frontend-engineer, backend-engineer, security-engineer, devops-engineer, performance-engineer, quality-engineer, mentor, refactorer, scribe, project-manager

10 coding agents: frontend-builder, backend-builder, debugger, reviewer, planner, tester, doc-writer, refactor-executor, code-explorer, code-architect

Skills (46)

  • Workflow (8): plan-before-coding, plan-mode, debug-systematically → systematic-investigation, test-before-commit, review-own-diff, incremental-refactor, ask-before-assuming, tdd
  • Quality & security (7): OWASP top 10, input validation, secrets hygiene, a11y basics, perf budget, api-design-reviewer, dependency-auditor
  • Tool-use (5): when to reach for Playwright, Context7, sequential-thinking, memory, semantic editing
  • Git (3): conventional commits, small focused PRs, PR description template
  • Framework-aware (16): Next.js, React, API patterns, FastAPI, Django, Flask, Prisma, Tailwind, TypeScript, pytest, and more — only loaded if detected
  • Meta (7): persona auto-activation, wave orchestration, ultracompressed mode, command-flags-spec, serena-semantic-editing, claude-md-improver, and more

Supported stacks (auto-detected)

  • Next.js (app router patterns, server actions, RSC guidance)
  • React (Vite, CRA — hooks rules, state colocation)
  • Node API (Express, Fastify, Hono, Koa — REST conventions, error handling)
  • Python API (FastAPI, Flask, Django — DI patterns, Pydantic)
  • Rust, Go — basic detection + generic guidance
  • Generic fallback — universal content for anything else

Detection is based on package.json, pyproject.toml, requirements.txt, go.mod, Cargo.toml.


Updating

npx clanker-code update

Performs an interactive 3-way merge. For each changed file, clanker compares your installed version against the original (shipped) version and the new (latest) version:

  • Untouched files → auto-updated
  • Your modifications to files we didn't change → kept as-is
  • New files → added
  • Conflicts (both you and we changed it) → prompt: keep yours / take new / skip

No destructive overwrites. Your changes are safe.


Configuration

All opt-outs via environment variables (no config file needed):

Variable Effect
CLANKER_HOOKS=off Disable all clanker hooks globally
CLANKER_HOOKS_EXTRA=on Enable opt-in hooks (session log, typecheck, pkg-install, transcript backup, ExitPlanMode auto-approve)
CLANKER_HOOK_<NAME>=off Disable a specific hook by name
CLANKER_NOTIFY_THRESHOLD_MS Override stop-notify threshold (default: 30000ms)
CLANKER_HOOK_CONTEXT_REINJECT_ALWAYS=on Re-inject on every SessionStart, not just compact
CLANKER_DEBUG=1 Show stack traces on CLI errors

Safety & trust

clanker-code is designed to be safe to run in any repository:

  • No network calls from hooks or CLI (except MCP servers which you explicitly enable)
  • No telemetry — nothing phones home
  • No auto-modify of your files beyond format-on-save and the content clanker installs
  • Merge-on-collision by default — never silently overwrites existing .claude/ or CLAUDE.md
  • Backup on replace — explicit backup-and-replace option preserves your previous config
  • Destructive commands blocked at the hook layer (rm -rf /, force-push to main, etc.)
  • Secret leak guard blocks writing obvious API keys
  • MIT licensed — open source, auditable

See SECURITY.md for the full security policy and vulnerability reporting.


FAQ

Does clanker-code work on Windows? Yes — native Windows, WSL, and Mac/Linux are all supported. The CLI is cross-platform Node.js with no native deps.

What if I already have a .claude/ folder? init detects it and prompts: merge (add missing clanker content, keep yours), backup-and-replace (saves existing to .clanker-backup/), or abort. Never overwrites silently.

How do I uninstall? Delete .claude/, .mcp.json, and the clanker-code section of CLAUDE.md (between the <!-- clanker-code:start --> / <!-- clanker-code:end --> markers).

Does Playwright really need 150MB? Only on first use — Playwright downloads Chromium when you run it the first time. You can pre-download with npx playwright install chromium, or skip Playwright entirely with npx clanker-code init --no-mcps and add what you want via mcp-help.

How do I add a team-specific skill or command? After init, .claude/skills/ and .claude/commands/ are yours to edit. Your changes persist through clanker update unless the file was shipped by clanker and you didn't modify it.

Can I use clanker-code in a private repo? Yes. The CLI reads only your project files locally. Nothing is uploaded.

Is clanker doctor a real command? Yes — it sanity-checks your Node version, shipped content, MCP servers, and common misconfigurations. Run it any time something feels off.


Contributing

See CONTRIBUTING.md. The governance rule is simple: every skill, agent, command, hook, and MCP must earn its slot. PRs that add content without justifying startup-token cost will be closed with a pointer to existing content.


License

MIT © Umayer


Links

About

The Claude Code starter kit. One command installs curated skills, agents, MCPs, hooks, and safety rails.

Resources

License

Contributing

Security policy

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors