A CLI-first multi-agent orchestration system for Claude Code. 13 specialized agents coordinate software development from research through deployment, enforced by hooks and guided by playbooks.
Prometheus turns Claude Code into a full development team:
/analyst -> /architect -> /pm -> /sm -> /dev -> /qa -> /po -> /devops
Research Design Plan Stories Code Test Accept Ship
Each agent owns its domain. Hooks enforce governance. Playbooks coordinate multi-agent workflows. Memory persists across sessions.
# 1. Clone and install
git clone https://github.com/igoroliveirg/prometheus-agent-system.git
cd prometheus-agent-system
./install.sh
# 2. Open Claude Code and initialize
/master *init # Bootstrap your project
/master *health # Verify system integrity
/help # Full command reference| Command | Agent | Domain |
|---|---|---|
/master |
Orchestrator | Cross-agent coordination, workflow management |
/analyst |
Analyst | Research, market analysis, project briefs |
/architect |
Architect | System design, tech stack, architecture docs (gatekeeper) |
/pm |
PM | PRDs, epics, product strategy |
/sm |
Scrum Master | Story creation, refinement, dependency tracking |
/po |
Product Owner | Story validation, acceptance, backlog management |
/dev |
Developer | Code implementation (story-bound) |
/qa |
QA | Testing, code review, quality gates |
/devops |
DevOps | Exclusive push/deploy authority, CI/CD |
/data-engineer |
Data Engineer | Database schemas, migrations, RLS, queries |
/ux |
UX | User flows, wireframes, design system, accessibility |
/squad |
Squad | Custom team formation for domains |
/loop |
Loop | Context continuity, state persistence, memory |
Three Python hooks run as Claude Code PreToolUse hooks:
| Hook | Rule | Effect |
|---|---|---|
enforce-architecture-first.py |
No files in src/ without matching arch doc in docs/architecture/ |
Blocks Write/Edit |
no-direct-push.py |
All git push blocked except via /devops *push |
Blocks Bash |
guardrails.py |
Lock files blocked, .env/config/migration warned, test ratio enforced |
Blocks/Warns |
Pre-defined multi-agent workflows. Run via /master *playbook [name]:
| Playbook | Agents | Use When |
|---|---|---|
new-feature |
all agents | Building something new |
bug-fix |
dev -> qa -> devops | Fixing a known bug |
hotfix |
dev -> qa -> devops (expedited) | Production emergency |
spike |
analyst -> architect | Research / proof of concept |
refactor |
architect -> dev -> qa -> devops | Restructuring code |
data-change |
architect -> data-eng -> dev -> qa -> devops | Schema/migration work |
ux-overhaul |
analyst -> ux -> architect -> dev -> qa -> po -> devops | UX redesign |
launch-prep |
qa -> devops -> po | Pre-release checklist |
retrospective |
dev -> qa -> architect -> sm -> master | Post-story learning |
warm-start |
loop -> master -> relevant agent | Resuming after inactivity |
Emergency protocols: production-down, security-breach, data-corruption, dependency-vulnerability
/master *status System-wide status
/master *kickoff [brief] Start new initiative
/master *playbook [name] Run a workflow
/master *health System integrity check
/master *who [capability] Find the right agent
/master *emergency [type] Incident response
/master *validate-docs Documentation consistency check
/master *backup Create system backup
/loop *save Checkpoint progress
/loop *resume Continue from last save
/loop *handoff [agent] Save + switch agent
/loop *compress Compress state for efficiency
/loop *rollback Restore previous state
/loop *remember [cat] [t] Write to shared memory
/loop *prompts [agent] Browse prompt library
/loop *summary Session changelog
/architect *design [feat] Create architecture doc
/sm *story [epic] [title] Create story from epic
/dev *implement [story] Implement a story
/qa *review [story] Code review
/po *accept [story] Accept completed work
/devops *push Push and create PR
Prometheus maintains context across sessions:
- State:
~/.claude/state/current.md-- Active session, current story, progress - Memory:
~/.claude/memory/-- Project knowledge (decisions, patterns, gotchas, stack, conventions) - Inbox:
~/.claude/inbox/-- Agent-to-agent async messaging - Audit Log:
~/.claude/logs/audit.md-- Append-only action trail - Metrics:
~/.claude/metrics/dashboard.md-- Performance tracking - Global Memory:
~/.claude/global-memory/-- Cross-project knowledge
Auto-save triggers every ~20 tool calls, before handoffs, and before risky operations.
Prometheus includes tools to keep itself consistent:
| Tool | Command | Purpose |
|---|---|---|
| Health Check | python3 ~/.claude/hooks/health-check.py |
Verify all 127+ system files |
| Docs Validator | python3 ~/.claude/hooks/docs-validator.py |
Catch documentation drift |
| Backup/Restore | python3 ~/.claude/hooks/backup.py backup |
Full system backup |
| Hook Tests | python3 ~/.claude/hooks/test-hooks.py |
18 integration tests |
When using Cursor instead of Claude Code CLI:
- Use
@agent-nameinstead of/agent-name - Copy
src/cursor-rules/to your project's.cursor/rules/ - Rules define agent authority, naming conventions, and governance
prometheus-agent-system/
├── src/ # All Prometheus source files
│ ├── CLAUDE.md # System constitution and quick reference
│ ├── VERSION # Current version (semver)
│ ├── CHANGELOG.md # Version history
│ ├── commands/ # 13 agent commands + context loader + help + PR template
│ ├── hooks/ # 7 Python scripts (3 enforcement + 4 utility)
│ ├── playbooks/ # 16 workflow recipes
│ ├── prompts/ # Reusable prompt snippets per agent
│ ├── profiles/ # 5 environment configs
│ ├── registry/ # Agent capabilities + skill map + plugin bridge
│ ├── guardrails/ # Safety rules + circuit breaker
│ ├── cursor-rules/ # Cursor IDE integration rules
│ ├── global-memory/ # Cross-project knowledge templates
│ ├── state/ # Session state template
│ ├── memory/ # Project knowledge templates
│ ├── inbox/ # Agent inbox templates
│ ├── logs/ # Audit log template
│ └── metrics/ # Dashboard template
├── templates/ # Project-level templates
│ └── stories/ # Story template + dependency graph
├── docs/ # Documentation
├── install.sh # Installer
├── uninstall.sh # Uninstaller
├── configure-hooks.py # Hook configuration tool
└── settings.template.json # Claude Code settings template
- Getting Started -- Installation, first project setup
- Agent Reference -- All 13 agents with full command lists
- Hook Reference -- How enforcement hooks work
- Playbook Guide -- Workflow recipes explained
- System Architecture -- Design decisions and system map
- Cursor Integration -- Using with Cursor IDE
- Claude Code CLI
- Python 3.8+
ghCLI (recommended, for/devops *push)
MIT
Current: v5.0.0 -- See CHANGELOG for history.