hool-cli
v0.9.4
Published
Agent-Driven SDLC — scaffold and configure HOOL for any project
Maintainers
Readme
hool-cli
Agent-Driven SDLC. Describe what you want to build, then step away. Agents handle spec, design, architecture, implementation, testing, and bug fixing.
Quick Start
npx hool-cli initYou'll be asked three questions:
- Platform — Claude Code, Cursor, or generic
- Project type — web app, browser game, mobile, CLI tool, API, desktop, animation, or other
- Mode — Interactive (you review spec/design/architecture) or Full-HOOL (fully autonomous)
Then open your AI coding tool and say:
Begin Phase 1: BrainstormInstall
npm install -g hool-cliUpdate to the latest version:
npm update -g hool-cliCommands
hool init
Scaffold HOOL in the current directory. Creates phase directories, operations files, agent memory, hooks, settings, skills, and platform instructions.
hool init
hool init --dir ./myapp --platform claude-code --type web-app --mode full-hoolOptions:
-d, --dir <path>— Project directory (default:.)-p, --platform <platform>—claude-code,cursor, orgeneric-t, --type <type>—web-app,browser-game,mobile-android,animation,cli-tool,api-only,desktop,other-m, --mode <mode>—interactive(default) orfull-hool
hool onboard
Onboard an existing codebase into HOOL. Scaffolds the full HOOL structure around your existing code (no files are touched), then sets up 11 onboarding tasks for agents to reverse-engineer project docs.
hool onboard
hool onboard --dir ./myapp --platform claude-code --type web-appIf .hool/ already exists, auto-detects and offers re-onboard — a lightweight path that only resets current-phase.md and prepends onboarding tasks to the task board, preserving all memory and phase docs.
hool status
Show current phase, task progress bars, per-agent breakdown, bug count, inconsistencies, governor audit status, and human review status.
hool statushool reset
Reset operations files and agent memory. Phase documents are preserved.
hool resethool mode [new-mode]
Show or switch execution mode.
hool mode # show current mode
hool mode full-hool # switch to full-hool
hool mode interactive # switch to interactiveModes
- Interactive (default) — You review and approve spec, design, and architecture (Phases 0-4) before agents build autonomously.
- Full-HOOL — You brainstorm the idea, then agents handle everything. Decisions are logged for post-build review.
Eight Agents
| Agent | Role | |-------|------| | Product Lead | Vision, contracts, doc consistency, phase gating, agent dispatch | | FE Tech Lead | FE scaffold, low-level design, code review | | BE Tech Lead | BE scaffold, low-level design, code review | | FE Dev | Frontend implementation | | BE Dev | Backend implementation | | QA | Test plan, test execution, bug reporting | | Forensic | Root cause analysis, bug triage, fix routing | | Governor | Behavioral auditor, rule enforcement, corrective feedback |
The Product Lead is the only user-facing agent. All others are dispatched as independent CLI sessions with their own context windows, MCP access, and role-specific hooks.
What Gets Created
your-project/
.hool/
phases/ # Phase docs (spec, design, architecture, LLDs, test plan)
operations/ # Live state (task board, bugs, issues, governor rules, logs)
memory/ # Per-agent memory (8 agents × 5 memory files each)
hooks/ # Platform hooks (context injection, governor triggers)
settings/ # Per-role settings (hooks, permissions)
checklists/ # Code review checklists
agents.json # Agent manifest
mcps.json # MCP manifest
.claude/
agents/ # Agent definitions (7 subagents)
skills/ # Phase skills (/brainstorm, /spec, /design, /architecture)
settings.json # Product Lead hooks
CLAUDE.md # Full orchestrator prompt (claude-code platform)Supported Platforms
| Platform | Instruction file | MCP support |
|----------|-----------------|-------------|
| Claude Code | CLAUDE.md | Full (auto-installs to ~/.claude/mcp_servers.json) |
| Cursor | .cursor/rules/hool.mdc | Full (auto-installs to ~/.cursor/mcp.json) |
| Generic | HOOL-INSTRUCTIONS.md | Manual setup |
Links
License
MIT
