@arthai/agents

AI development toolkit for Claude Code — agents, skills, and hooks organized into installable bundles.

Quick Start

# 1. Get a license key (email [email protected])

# 2. Activate your license (one time)
npx @arthai/agents activate ARTH-XXXX-XXXX-XXXX-XXXX

# 3. Install a bundle into your project
npx @arthai/agents install forge .

# 4. Open Claude Code — skills are ready
/calibrate                  # auto-configure for your codebase
/planning my-feature        # start building

Don't have a key? Email [email protected] to get one.

Already have the toolkit cloned at ~/.claude-agents/? You don't need to activate — your existing key works automatically.

Explore Before Installing

No license needed to browse:

# See all available bundles with descriptions
npx @arthai/agents list

# See what's inside a specific bundle (agents, skills, hooks)
npx @arthai/agents info forge

# See what's inside every bundle
npx @arthai/agents info prime

Install

# Install a bundle into your project
npx @arthai/agents install forge .

# Install multiple bundles at once
npx @arthai/agents install forge spark shield .

# Install everything
npx @arthai/agents install prime .

What gets installed

When you install a bundle, these files are created in your project:

your-project/
  .claude/
    skills/          # slash commands (/planning, /implement, /qa, etc.)
    agents/          # specialized AI agents (architect, backend, QA, etc.)
    hooks/           # automation hooks (guardrails, deploy health, etc.)
    settings.json    # hooks registered here
  CLAUDE.md          # project context file (created if missing)
  .gitignore         # updated with toolkit entries

How skills, agents, and hooks work

| Component | What it is | How you use it | Example | |-----------|-----------|----------------|---------| | Skills | Slash commands with structured workflows | Type /skillname in Claude Code | /planning my-feature — spawns PM + Architect + DA for adversarial planning | | Agents | Specialized AI personas with domain expertise | Skills spawn them automatically, or use Agent tool directly | architect — designs systems, backend — writes backend code, qa — validates | | Hooks | Shell scripts that fire on events | Automatic — run on session start, before/after commands | pre-bash-guard — blocks dangerous commands, escalation-guard — circuit breaker on 3 failures |

Skills are what you interact with. Agents and hooks work behind the scenes.

Bundles

Pick what you need. Each bundle is a self-contained set of agents, skills, and hooks.

Development

| Bundle | What you get | |--------|-------------| | forge | Full dev workflow: /planning, /implement, /qa, /pr, /precheck, /review-pr + architect, PM, backend, frontend, QA, code-reviewer agents + triage router + escalation guard | | scalpel | Bug fixing: /fix, /ci-fix, /issue + code-reviewer + troubleshooter agents | | spark | Project setup: /onboard, /calibrate, /scan, /setup + setup and explore agents |

Operations

| Bundle | What you get | |--------|-------------| | sentinel | SRE + ops: /sre, /incident, /restart + SRE, ops agents + deploy health, port guard, crash watcher hooks | | prism | Deep QA: /qa, /qa-learn, /qa-incident + 6 QA agents (qa, e2e, domain, challenger, baseline, promoter) | | shield | Safety guardrails only: bash guard, edit guard, session bootstrap, triage router, escalation guard (no skills — hooks only) |

Strategy

| Bundle | What you get | |--------|-------------| | canvas | Design: /planning + design-studio agents (think, create, critique) + frontend agent | | compass | Product: /planning + PM, GTM, user researcher, content strategist agents | | counsel | Consulting: 13 skills (client-discovery, proposals, deliverables, ROI, etc.) + 4 consulting agents |

Power

| Bundle | What you get | |--------|-------------| | cruise | Autonomous mode: /autopilot — picks up issues, implements, QAs, creates PRs. Auto-installs forge + scalpel + sentinel. | | prime | Everything. All agents, skills, and hooks. |

Which bundle should I start with?

| You want to... | Install | |----------------|---------| | Build features end-to-end | forge | | Fix bugs with formal pipeline | scalpel | | Get set up on a new project | spark | | All of the above | forge spark scalpel | | Everything including SRE, QA, design, consulting | prime |

Workflows

After installing, use skills in Claude Code:

Start a new project:

/onboard                    # get project briefing
/calibrate                  # auto-configure toolkit for your codebase
/planning my-feature        # design the feature with architect + PM agents
/implement my-feature       # build with parallel agent team

Fix a bug:

/fix #42                    # root cause analysis + fix + regression tests
/ci-fix                     # auto-repair CI failures

Ship code:

/precheck                   # local CI in 30s
/qa commit                  # targeted QA on changed files
/pr                         # create PR with QA results

Autonomous mode:

/autopilot                  # picks up issues, implements, QAs, creates PRs

Operations:

/sre status                 # health check all services
/incident                   # triage + parallel diagnosis
/restart                    # restart local dev servers

Updating

# Update to latest version (re-run install with @latest)
npx @arthai/agents@latest install forge .

When to update

• After a new release — check the changelog or Discord #deployments
• When a skill isn't working as expected — updates often fix edge cases
• After npx cache issues — clear cache and reinstall:

# Clear npx cache if you're getting stale versions
rm -rf ~/.npm/_npx
npx @arthai/agents@latest install forge .

What happens on update

• Skills, agents, and hooks are overwritten with the latest versions
• Your CLAUDE.md is never overwritten — only the managed toolkit block is updated
• Your .claude/settings.json hooks are merged — existing hooks are preserved
• Your project-specific files (.claude/project-profile.md, .claude/knowledge/) are never touched

Uninstall

# Remove a specific bundle
npx @arthai/agents uninstall forge .

# Remove all bundles at once
npx @arthai/agents uninstall prime .

This removes skills and agents (63 files for prime). To fully clean up including hooks:

# 1. Remove hook scripts
rm -rf .claude/hooks

# 2. Remove hook entries from settings.json
python3 -c "
import json
with open('.claude/settings.json') as f: s = json.load(f)
s.pop('hooks', None)
with open('.claude/settings.json', 'w') as f: json.dump(s, f, indent=2)
"

What uninstall removes vs keeps:

| Removed | Kept (never deleted) | |---------|------| | .claude/skills/ | CLAUDE.md (your project context) | | .claude/agents/ | .claude/project-profile.md (calibration data) | | | .claude/knowledge/ (project knowledge base) | | | .claude/settings.json (hook entries — clean manually) | | | .claude/hooks/ (script files — rm -rf manually) | | | ~/.arthai/license (license key — per-machine) |

License Activation

A license key is required to install bundles. Browsing (list, info) is free.

# Activate (one time — saves key locally)
npx @arthai/agents activate ARTH-XXXX-XXXX-XXXX-XXXX

# Key is stored at ~/.arthai/license — all future installs use it automatically

For CI/teams: Set ARTHAI_LICENSE_KEY environment variable instead of activating per-machine.

Key precedence: ARTHAI_LICENSE_KEY env var > ~/.arthai/license file > --key flag

Troubleshooting

| Problem | Fix | |---------|-----| | Skills not showing in Claude Code | Restart Claude Code in the project directory | | /planning says unknown skill | Verify: ls .claude/skills/planning/SKILL.md | | Install crashes on CLAUDE.md setup | Update: rm -rf ~/.npm/_npx && npx @arthai/agents@latest install forge . | | "License required" after activation | Check: cat ~/.arthai/license — should contain your key | | "License server unavailable" | Check internet connection. Try again in a minute. | | Old version stuck | Clear cache: rm -rf ~/.npm/_npx then reinstall | | Hooks not firing | Check .claude/settings.json has hook entries. Restart Claude Code. |

Best Practices

1. Start with forge — it covers 80% of workflows. Add other bundles as needed.
2. Run /calibrate after install — this teaches the toolkit your project's patterns, stack, and conventions. Without it, agents work generically.
3. Run /onboard every session — quick briefing on open PRs, issues, and priorities.
4. Use /precheck before pushing — catches CI failures locally in 30s instead of a 4-minute round-trip.
5. Don't edit files in .claude/skills/ or .claude/agents/ — they get overwritten on update. Customize by adding your own files alongside them.
6. Keep CLAUDE.md up to date — agents read it for project context. The more accurate it is, the better they perform.

Gotchas

• npx caches aggressively — if you're not getting the latest version, run rm -rf ~/.npm/_npx before reinstalling
• Hooks stack, they don't replace — installing a bundle twice adds duplicate hook entries to settings.json. Uninstall first if reinstalling.
• CLAUDE.md managed block is auto-updated — the section between <!-- >>> claude-agents toolkit --> markers is overwritten on install. Don't put your content there.
• Agent teams require experimental flag — set CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1 in your Claude Code settings (auto-configured by the installer)

Also available as Claude Code plugin

/plugin marketplace add ArthTech-AI/arthai-marketplace
/plugin install forge@arthai-marketplace

Requirements

• Claude Code CLI
• Node.js 18+
• License key (get one at [email protected])