smash-os-install
v0.4.6
Published
Install the SmashOS local AI workflow harness into any repo — no web app, no API keys required
Maintainers
Readme
smash-os-install
Install the SmashOS local AI workflow harness into any repository. No web app, no API keys, no cloud dependencies.
npx smash-os-installWhat SmashOS is
SmashOS gives Claude Code a coordinated team of 12 AI specialists — each with a defined role, scoring rules, and memory — that work together through structured pipelines to ship features, catch bugs, and audit security. Everything runs locally via Claude Code with no external services.
Install
# Run inside your repo root
npx smash-os-installPrerequisites: Node.js 18+ and Claude Code
The installer asks for your project name and tech stack, then writes all harness files into your repo.
What gets created
your-repo/
├── CLAUDE.md ← harness instructions loaded by Claude on every session
├── .claude/
│ ├── hooks/
│ │ ├── boot-hook.mjs ← injects context files at session start
│ │ └── sync-hook.mjs ← persists memory after each response
│ └── agents/ ← 47 sub-agent definitions
├── ai/
│ ├── context/
│ │ ├── product.md ← fill in: what your product does
│ │ ├── architecture.md ← fill in: your stack and key decisions
│ │ └── coding-standards.md ← fill in: your style rules
│ ├── roles/ ← 12 role definitions (PM, Staff Eng, Dev, Security, QA, DevOps…)
│ ├── workflows/ ← 10 pipeline definitions
│ └── memory/
│ ├── decisions.md ← auto-populated: architectural decisions
│ ├── lessons.md ← auto-populated: what went wrong + why
│ └── constraints.md ← auto-populated: hard limits
└── .claude-state/ ← session breadcrumbs (gitignored)
├── active-context.md
├── next-steps.md
└── decisions.mdSkills install globally to ~/.claude/skills/ and are available in every project.
First 3 commands
# 1. Open Claude Code in your repo
claude .
# 2. Verify the installation
/smash-os:status
# 3. Run your first pipeline
/smash-os:run feature "Add a health check endpoint"How a pipeline works
/smash-os:run feature "your task"
│
├─ Phase 1: Product Manager → spec + acceptance criteria (score ≥ 60 to pass)
├─ Phase 2: Staff Engineer → architecture review + spec sign-off
├─ Phase 3: Senior Developer → implementation (branch + commits)
├─ Phase 4: Security Engineer → threat model + security_score
├─ Phase 5: QA Engineer → test plan + test_success_score
└─ Phase 6: DevOps Engineer → PR + deployment readinessEach phase scores its output 0–100. Below 60 halts the pipeline. The blocking reason is written to ai/memory/ and injected into the next session automatically.
Architecture
┌─────────────────────────────────────────────────────┐
│ Claude Code │
│ │
│ CLAUDE.md ──→ loads harness instructions │
│ boot-hook ──→ injects ai/context/ + ai/memory/ │
│ sync-hook ──→ writes decisions + lessons back │
│ │
│ /smash-os:run feature "task" │
│ │ │
│ └─→ Pipeline orchestrator │
│ │ │
│ ├─→ Role loader (ai/roles/) │
│ ├─→ Workflow runner (ai/workflows/) │
│ └─→ Sub-agents (.claude/agents/) │
│ │ │
│ └─→ Phase outputs → memory │
└─────────────────────────────────────────────────────┘Slash commands
| Command | Description |
|---------|-------------|
| /smash-os:status | Health check — active pipelines, recent decisions, memory state |
| /smash-os:run [type] | Start a pipeline: feature, bug-fix, security-audit, weekly-improvement |
| /smash-os:fast | Skip the pipeline — execute a trivial task inline (≤3 files) |
| /smash-os:role [name] | Switch to a specific role |
| /smash-os:memory | Show stored decisions, lessons, constraints |
| /smash-os:audit | Trigger an immediate security audit |
| /smash-os:doctor | Verify the installation is complete and correct |
| /smash-os:evolve | Run harness self-improvement on accumulated role learnings |
| /smash-os:sync | Manual mid-session checkpoint — save current state to memory |
License
MIT
