Why Claude Hangar?
Most Claude Code configs are personal dotfiles — useful to read, hard to reuse. Claude Hangar is different:
- One command deploys a complete, tested setup to
~/.claude/ - Multi-project orchestration — manage configs for multiple repos from one place
- Modular stacks — pick only what you need (
bash integrate.sh <stack>) with--with/--withoutcomponent control - Battle-tested hooks that prevent real incidents (secret leaks, destructive commands, context overflow)
- 42 skills from project scanning to pre-PR verification, context optimization, session recap, effort tuning, stack lenses, and meta-governance
- 31 lifecycle hooks with 4-level quality gates, config protection, MCP health checks, design quality guards, permission-request inspection, budget caps, and smart context preservation
- 22 agents for specialized tasks (code review, TDD enforcement, performance analysis, MCP server development, autonomous workflows) — deep-reasoning agents run at
effort: xhigh(v2.1.111+) - 20 governance rules (common + language-specific) always-on code quality
- 3 context modes — dev, research, review
- Cross-platform — Linux, macOS, and Windows (Git Bash)
Prerequisites
- Node.js >= 18 LTS (for hook JSON parsing via
node -e) - Bash >= 4.0 (Git Bash on Windows, native on macOS/Linux)
- Git >= 2.20
- GitHub CLI (
gh) — optional, for GitHub stack features and freshness checks
Quick Start
One-liner install:
curl -fsSL https://raw.githubusercontent.com/claude-hangar/claude-hangar/main/install.sh | bash
Or clone manually:
git clone https://github.com/claude-hangar/claude-hangar.git ~/.claude-hangar
cd ~/.claude-hangar && bash setup.sh
Claude Hangar — Setup
─────────────────────
[i] Running on: linux
[+] Prerequisites: git ✓ node ✓
[+] Structure validation passed
[i] Deploying to /home/user/.claude/...
[+] Deployed: Hooks (31 scripts)
[+] Deployed: Agents (22 definitions)
[+] Deployed: Skills (41 commands)
[+] Deployed: Shared lib
[+] Deployed: Statusline
[+] Deployed: settings.json (from template)
[+] Deployment complete!
[i] Open Claude Code in any project to start.
What You Get
Hooks (27) — Automated Safety Net
| Hook | Event | What It Does |
|---|---|---|
secret-leak-check |
PreToolUse | Blocks writes containing API keys, tokens, or credentials |
bash-guard |
PreToolUse | Prevents destructive commands (rm -rf, DROP TABLE, force-push) |
checkpoint |
PreToolUse | Auto-creates git stash snapshots before file edits |
config-protection |
PreToolUse | Blocks weakening of linter/formatter/compiler configs |
mcp-health-check |
PreToolUse | Checks MCP server health before tool calls, warns on repeated failures |
db-query-guard |
PreToolUse | Warns when agent tries to directly query internal databases |
permission-denied-retry |
PermissionDenied | Retries with adjusted approach on permission denial |
token-warning |
PostToolUse | Alerts at 70% and 80% context utilization |
continuous-learning |
PostToolUse | Captures patterns and learnings from agent work |
cost-tracker |
PostToolUse | Tracks token costs per session |
design-quality-check |
PostToolUse | Detects generic AI UI drift patterns in frontend files |
batch-format-collector |
PostToolUse | Collects edited file paths for batch formatting |
instinct-capture |
PostToolUse | Captures instinct patterns from tool usage |
skill-suggest |
UserPromptSubmit | Suggests matching skills based on your prompts |
model-router |
UserPromptSubmit | Smart complexity analysis — structural signals, scope detection |
session-start |
SessionStart | Loads STATUS.md, tasks, and memory on session start |
config-change-guard |
ConfigChange | Warns on critical settings changes |
task-completed-gate |
TaskCompleted | 4-level quality gate (existence, errors, evidence, substance) |
task-created-init |
TaskCreated | Initializes new tasks with metadata |
subagent-tracker |
SubagentStart/Stop | Lifecycle tracking + forensics (duration, thrashing, failures) |
post-compact |
PostCompact | Smart context preservation — detects tasks, plans, branch, HANDOFF.md |
worktree-init |
WorktreeCreate | Initializes worktree environment for isolated work |
session-stop |
Stop | Cleans temp files, logs session cost |
stop-failure |
StopFailure | Logs errors on session failures |
stop-batch-format |
Stop | Runs formatters once at session end on all edited files |
instinct-evolve |
Stop | Evolves instinct data from session patterns |
desktop-notify |
Stop | OS notification when session ends (Windows/Linux/macOS) |
MCP Server — Hangar State API
A zero-dependency MCP server (core/mcp-server/server.js) that exposes Hangar configuration state as 5 read-only tools: hangar_hooks, hangar_skills, hangar_agents, hangar_config, hangar_freshness.
Shared References (3)
Shared behavioral files that skills and agents reference for consistency:
| Reference | Purpose |
|---|---|
questioning.md |
Shared questioning style for skills and agents |
design-principles.md |
Design quality standards (no AI aesthetic, WCAG AA, curated palettes) |
code-quality.md |
Code quality defaults (fix code not configs, immutability, coverage) |
Agents (21) — Specialized AI Workers
| Agent | Model | Purpose |
|---|---|---|
explorer |
Sonnet | Fast read-only codebase analysis |
explorer-deep |
Opus | Deep architecture analysis (read-only) |
security-reviewer |
Opus | OWASP Top 10 + Agentic Top 10 security audit |
commit-reviewer |
Sonnet | Pre-commit review for staged changes |
plan-reviewer |
Sonnet | Spec/plan compliance — verifies nothing more, nothing less |
dependency-checker |
Sonnet | npm audit + outdated packages + CVE research |
planner |
Opus | Complex feature implementation planning |
architect |
Opus | System design and architecture decisions |
loop-operator |
Sonnet | Autonomous workflow with safety guardrails |
tdd-guide |
Sonnet | TDD enforcement (RED-GREEN-REFACTOR cycle) |
doc-updater |
Sonnet | Documentation maintenance and staleness detection |
refactor-agent |
Opus | Code restructuring and cleanup (worktree isolation) |
test-writer |
Sonnet | Test generation for new features |
typescript-reviewer |
Sonnet | TypeScript-specific code review |
python-reviewer |
Sonnet | Python-specific code review |
go-reviewer |
Sonnet | Go-specific code review |
build-resolver-typescript |
Sonnet | TS/JS build error resolution |
build-resolver-python |
Sonnet | Python build error resolution |
build-resolver-go |
Sonnet | Go build error resolution |
harness-optimizer |
Opus | Self-optimization of hooks, skills, rules, and agent configs |
performance-optimizer |
Opus | Bundle analysis, profiling, Core Web Vitals, memory leaks |
Skills (31) — Slash Commands for Real Work
| Skill | Description |
|---|---|
/scan |
Auto-detect tech stack, architecture, and generate CLAUDE.md |
/consult |
Interactive improvement wizard with structured plans |
/verify |
Pre-PR 6-phase quality pipeline (build, types, lint, test, security, diff) |
/context-budget |
Token spending analysis and optimization opportunities |
/strategic-compact |
Smart /compact timing based on workflow state |
/audit |
Three-layer website audit (9 phases) |
/project-audit |
Repository audit (10 phases) |
/adversarial-review |
Critical review — minimum 5 findings |
/safety-guard |
3-mode write protection (Careful/Freeze/Guard) for autonomous runs |
/rules-distill |
Extract cross-cutting rules from skill collection |
/skill-stocktake |
Audit skill quality across 4 dimensions |
/prompt-optimizer |
6-phase prompt analysis and skill matching pipeline |
/polish |
Frontend quick wins across 6 dimensions |
/deploy-check |
Docker/Traefik deployment readiness |
/git-hygiene |
Stale branches, large files, commit quality |
/freshness-check |
Framework and dependency version tracking |
/lighthouse-quick |
Core Web Vitals performance check |
/capture-pdf |
Multi-page website PDF capture |
/meta-tags |
OG/Twitter/Structured Data validation |
/favicon-check |
Icon completeness check |
/design-system |
Tailwind v4 design reference |
/lesson-learned |
Extract and persist learnings |
/audit-orchestrator |
Multi-audit coordination |
/audit-runner |
Autonomous batch audit execution |
/gsd-orchestrate |
GSD v1 end-to-end automation (checkpoint/full-auto/assisted, batch overnight runs) |
/handoff |
Structured session handoff for seamless context continuity |
/codebase-map |
Structural overview for context recovery after /compact |
/doctor |
Project health check (git-hygiene + deploy-check + freshness) |
/error-analyzer |
Systematic root-cause analysis for build/test/runtime failures |
/inline-review |
Quick self-review checklist after code changes |
/pattern-extractor |
Analyze captured patterns from continuous learning |
/security-scan |
Claude Code security scan (secrets, MCP, hooks, deps) |
Stacks (9) — Framework & Tool Extensions
Drop-in extensions for your specific tech stack. Full stacks provide CLAUDE.md.snippet + audit skills. MCP stacks provide tool integrations.
| Stack | Type | Includes |
|---|---|---|
| Astro | Full | SSG/SSR patterns, content collections, View Transitions, /astro-audit |
| SvelteKit | Full | Svelte 5 runes, load functions, form actions, /sveltekit-audit |
| Next.js | Full | App Router, Server Components, Server Actions |
| Database | Full | Drizzle ORM, migrations, schema design, /db-audit |
| Auth | Full | Custom bcrypt + sessions, secure cookies, /auth-audit |
| Docker | Snippet | Docker/container CLAUDE.md context |
| GitHub | MCP | GitHub repos, PRs, issues via MCP server |
| Web | MCP | Browser automation (Playwright) via MCP server |
| Security | MCP | Security scanning (Snyk) via MCP server |
Templates
- 5 project templates — CLAUDE.md starters (minimal, web, fullstack, management) + DECISIONS.md (ADR register)
- 6 CI/CD templates — GitHub Actions for Node.js, Python, VPS (GHCR), Docker Compose, GitHub Pages, Cloudflare Pages
- Global CLAUDE.md template — Baseline instructions with
{{PLACEHOLDER}}customization
Multi-Project Orchestration
Manage Claude Code configs for multiple repositories from one registry.json:
{
"projects": [
{
"name": "website",
"repo": "my-org/website",
"defaultPath": "~/projects/website",
"skills": ["audit", "astro"],
"workflows": ["ci-node.yml", "deploy-ghpages.yml"]
},
{
"name": "api",
"repo": "my-org/api",
"defaultPath": "~/projects/api",
"skills": ["audit", "database", "auth"],
"workflows": ["ci-node.yml", "deploy-vps-ghcr.yml"]
}
]
}
Each project gets the right CLAUDE.md, stack-specific skills, and CI templates — all from one source of truth.
Repository Structure
| Directory | Purpose |
|---|---|
core/ |
Global config deployed to ~/.claude/ (hooks, agents, skills, lib, statusline) |
stacks/ |
Framework-specific extensions (Astro, SvelteKit, Next.js, Database, Auth) |
templates/ |
CI/CD workflows and project CLAUDE.md templates |
registry/ |
Multi-project management schema and examples |
tests/ |
Hook tests, setup tests, template tests |
docs/ |
Documentation, concepts, tutorials |
i18n/ |
Internationalization (currently: German) |
rules/ |
Always-on governance rules (common + language-specific) |
core/contexts/ |
Dynamic context modes (dev, research, review) |
Documentation
| Guide | Description |
|---|---|
| Getting Started | Installation and first run |
| Configuration | Settings, hooks, environment variables |
| Multi-Project Setup | Registry-based project management |
| Writing Skills | Create your own slash commands |
| Writing Hooks | Extend the hook system |
| Writing Agents | Custom agent definitions |
| Architecture | System design and internals |
| Companion Tools | Superpowers, Trail of Bits, ccusage, claude-squad |
| FAQ | Common questions |
Works Great With
Claude Hangar is the infrastructure layer. These companion tools extend it:
| Tool | Stars | What It Adds |
|---|---|---|
| Superpowers | 136K+ | Deep workflow methodology — brainstorming, TDD, subagent-driven development, systematic debugging |
| Trail of Bits Skills | 3.8K+ | Professional security skills — CodeQL, Semgrep, variant analysis, fix verification |
| ccusage | 11.8K+ | Historical usage analytics — token costs, session history, dashboards |
| claude-squad | 6.4K+ | Multi-session management — run multiple Claude Code instances in parallel |
| claude-mem | 39K+ | Persistent memory plugin — session captures, AI compression, semantic search |
| Everything Claude Code | 140K+ | Comprehensive system with instincts, memory, and security patterns |
All are compatible with Hangar and each other. No conflicts, no overlap.
Contributing
Contributions are welcome! See CONTRIBUTING.md for guidelines.
Areas where contributions are especially valued:
- New stacks — Rails, Django, Go, Rust, Laravel, ...
- Skills — Share your workflow automations
- Translations — Help us reach more developers (i18n guide)
- Bug reports — Issues with specific OS/shell combinations
Inspired By
- Everything Claude Code — Rules system, language agents, learning mechanisms, context modes, hook profiles, config protection, meta-governance, shared references
- GSD v2 — DECISIONS.md (append-only ADR register) pattern
- BMAD-METHOD — Story continuity, intent cascade patterns
- spec-kit — User-invocable frontmatter, argument hints
- oh-my-opencode — Agent fallback, hook composition patterns
- ui-ux-pro-max-skill — Design token architecture, product palettes
License
MIT — Use it, fork it, extend it.
No comments yet
Be the first to share your take.