Persona-based AI skill ecosystem for Claude Code, Claude Chat, and Agent SDK
Quick Start • Skill Tree • Personas • Install
AI assistants have great primitives (file access, shell, MCPs) but zero orchestration. You can enrich a lead with Clay, draft an email with Gmail, and schedule a follow-up — but nothing chains them into a single command.
jitsu-skills fills the gap: 14 orchestration skills organized by 7 personas, each combining multiple tools into end-to-end workflows. Universal format — works in Claude Code, Claude Chat, and any agent harness.
"Skills are just folders. Folders that teach Claude your job. Your workflow. Your expertise. Your domain. Claude on day 30 is a completely different tool than day one." — Barry Zhang & Mahesh Sathiamoorthy, Anthropic Engineers (source)
Full environment bootstrap (new device):
# One command — installs 200+ community skills, 14 orchestration skills,
# jitsu-os persona system, CLAUDE.md, hooks, and weekly auto-updater
curl -sL https://raw.githubusercontent.com/0xjitsu/jitsu-skills/main/scripts/bootstrap.sh | bashOr install just the orchestration skills:
# 1. Claude Code plugin (recommended)
claude plugin add jitsu-skills
# 2. npm (one-shot or global)
npx @0xjitsu/skills install
# 3. Manual (clone + symlink)
git clone https://github.com/0xjitsu/jitsu-skills.git
cd jitsu-skills && bun bin/cli.js install ── Core ──
★ Claude Core
── Tier 1 (unlocked) ──
📤 Template Personalizer ← core
📤 Lead Scorer ← core
🚀 Changelog Generator ← core
🚀 PII Sweep ← core
🛡️ Secret Scanner ← core
🛡️ Dependency Audit ← core
⚙️ Keychain Manager ← core
⚙️ LaunchAgent Builder ← core
🏗️ Supabase Schema ← core
🏗️ Chart Builder ← core
── Tier 2 (requires prerequisites) ──
📤 Campaign Sequencer ← template-personalizer + lead-scorer
🚀 npm Publish ← changelog-gen + pii-sweep
🔬 Proposal Builder ← chart-builder
🤝 Meeting Prep ← lead-scorer
Run npx @0xjitsu/skills tree for the interactive version.
Each skill belongs to a persona — a role the AI assumes when executing the skill.
| Persona | Icon | Skills | Domain |
|---|---|---|---|
| Campaign Ops | 📤 | campaign-sequencer, template-personalizer, lead-scorer | Outbound campaigns, lead enrichment, multi-channel sequencing |
| Ship Captain | 🚀 | npm-publish, changelog-gen, pii-sweep | Open source publishing, releases, pre-publish security |
| Guardian | 🛡️ | secret-scanner, dep-audit | Security auditing, vulnerability detection, git history scanning |
| Machinist | ⚙️ | launchagent-builder, keychain-manager | macOS automation, LaunchAgents, Keychain secret management |
| Architect | 🏗️ | supabase-schema, chart-builder | Full-stack dev, database design, data visualization |
| Researcher | 🔬 | proposal-builder | R&D proposals, citation management, single-page research apps |
| Coordinator | 🤝 | meeting-prep | Meeting briefings, cross-tool context aggregation |
📤 Campaign Ops
Multi-channel drip sequence orchestrator. Chains Clay enrichment → Gmail drafts → Slack DMs → Scheduled Tasks into a stateful pipeline with contact stage tracking.
Merge-field personalization engine with tone controls (formal, casual, CT-style). Pulls enrichment data from Clay and generates personalized message variants.
Weighted scoring from Clay enrichment data. Dimensions: company fit (30%), title seniority (25%), tech stack (20%), social presence (15%), engagement (10%).
🚀 Ship Captain
Full publish pipeline: version bump, CHANGELOG generation, build, pack dry-run, npm publish, GitHub release, git tag.
Parse conventional commits since last tag into categorized CHANGELOG.md entries (Added, Fixed, Changed, Breaking).
Pre-publish PII scrubbing: grep for emails, phone numbers, API keys, internal URLs. Replace with sample data. Verify with second pass.
🛡️ Guardian
Scan full git history (not just HEAD) for committed secrets. Flags commits, suggests git filter-repo remediation.
npm audit + Sonatype cross-reference for actionable vulnerability reports with fix commands.
⚙️ Machinist
Generate validated com.berna.* plist files for macOS LaunchAgents. Template-driven scheduling, XML validation, launchctl management.
Unified macOS Keychain interface. Store, retrieve, list, and rotate secrets via security CLI with naming conventions.
🏗️ Architect
Natural language to complete Supabase migration: tables, RLS policies, indexes, foreign keys, TypeScript types.
Dark-themed Chart.js/Recharts with glass morphism, accessibility (aria-labels, reduced-motion), responsive design.
🔬 Researcher
Scaffold single-page R&D proposal apps with Next.js, glass morphism, scroll-spy, citation system, and Lighthouse 100 checklist.
🤝 Coordinator
Pull Calendar event → enrich attendees via Clay/Slack → fetch Notion docs → compile briefing document.
Every skill works across platforms:
| Platform | How |
|---|---|
| Claude Code | Native — SKILL.md with frontmatter, auto-triggers |
| Claude Chat | Paste prompt.txt into Project Instructions |
| Agent SDK | Inject prompt.txt as systemPrompt string |
| npm | npx @0xjitsu/skills install symlinks into ~/.claude/skills/ |
Run bash scripts/build-universal.sh to generate prompt.txt from all skills.
npx @0xjitsu/skills list # Show all skills + install status
npx @0xjitsu/skills install # Symlink skills into ~/.claude/skills/
npx @0xjitsu/skills uninstall # Remove symlinked skills
npx @0xjitsu/skills tree # ASCII skill tree with prerequisitesjitsu-skills/
├── plugin.json # Claude Code plugin manifest
├── package.json # npm package config (@0xjitsu/skills)
├── bin/cli.js # CLI (list, install, uninstall, tree)
├── scripts/
│ └── build-universal.sh # Generate prompt.txt from SKILL.md
├── skills/
│ ├── skill-data.json # Skill tree graph (nodes, edges, personas)
│ ├── campaign-sequencer/ # 📤 Multi-channel drip sequences
│ ├── template-personalizer/ # 📤 Merge-field personalization
│ ├── lead-scorer/ # 📤 Weighted lead scoring
│ ├── npm-publish/ # 🚀 Full npm publish pipeline
│ ├── changelog-gen/ # 🚀 Conventional commit changelogs
│ ├── pii-sweep/ # 🚀 Pre-publish PII scrubbing
│ ├── secret-scanner/ # 🛡️ Git history secret detection
│ ├── dep-audit/ # 🛡️ Dependency vulnerability scanning
│ ├── launchagent-builder/ # ⚙️ macOS LaunchAgent generator
│ ├── keychain-manager/ # ⚙️ macOS Keychain manager
│ ├── supabase-schema/ # 🏗️ Natural language → Supabase migration
│ ├── chart-builder/ # 🏗️ Dark-themed data visualization
│ ├── proposal-builder/ # 🔬 R&D proposal app scaffolder
│ └── meeting-prep/ # 🤝 Multi-source meeting briefings
└── app/skill-tree/ # Interactive React Flow visualization
Open source under AGPL-3.0 for personal use and community projects. Commercial license available for teams, enterprises, and SaaS products. See LICENSE for details.
This repo is designed for both human and AI contributors.
| Resource | Purpose |
|---|---|
| CLAUDE.md | Full context for Claude Code sessions |
| llms.txt | Structured manifest for AI agent discovery |
| docs/adding-a-skill.md | Step-by-step skill creation guide |
| docs/architecture.md | How the universal skill format works |
prompt.txt files |
Portable skill definitions for any agent harness |
coding-discipline overlay |
Karpathy's four LLM-coding rules — applied on top of any persona when writing or editing code in this repo |
To contribute as an agent: read CLAUDE.md, layer on the coding-discipline overlay (Think Before Coding · Simplicity First · Surgical Changes · Goal-Driven Execution), create a skill following the guide, run bash scripts/build-universal.sh, submit a PR.
jitsu-skills — orchestration skills for the agent era
Get Started •
Report a Bug •
Request a Feature