Version: 2.0.0 Status: Active Development Built on: OpenClaw + Lobster + Community Skills
PM-Bot is an AI-powered project management bot that handles:
- SDLC: TDD-first development from stories to deployment with automated quality & security gates
- TDLC: Backtesting, strategy optimization, deployment
- RCA: Root cause analysis, log analysis, error tracing
- DevOps: Infrastructure management, deployments, monitoring
Discord User → OpenClaw (conversational AI)
↓
Lobster Pipelines (workflow orchestration)
↓
Community Skills (80%) + Custom PM-Bot Skills (20%)
↓
MCP Servers (filesystem, postgres, github, etc.)
Key Components:
- OpenClaw: Conversational AI framework with persistent memory
- Lobster: Typed workflow pipelines with approval gates
- llm-council: Multi-model orchestration (Claude + Gemini + GPT-4)
- pr-commit-workflow: GitHub automation
- tdd-guide: Test-driven development
- senior-devops: Infrastructure management
- OpenClaw installed
- Discord bot token (already configured)
- Git, Docker, Node.js
# 1. Clone or use existing /opt/pm-bot/
cd /opt/pm-bot
# 2. Configure environment
cp .env.example .env
# Edit .env and add your DISCORD_USER_ID
# 3. Setup OpenClaw
./scripts/setup-openclaw.sh
# 4. Install skills
./scripts/install-skills.sh
# 5. Start gateway
openclaw gateway startIn Discord:
# Multi-model planning
Plan the authentication refactor
# Full SDLC pipeline (original)
Run sdlc-full "Add health check endpoint"
# TDD-first SDLC pipeline (NEW)
Run sdlc-tdd-full "Add user profile API"
# TDD SDLC with conversational discovery (Phase 0)
Run sdlc-with-discovery "Make user profiles better"
# Root cause analysis
Why is the API returning 500 errors?
# Scheduled monitoring
Every Monday at 9am, summarize open PRs
/opt/pm-bot/
├── docs/ # Documentation
│ ├── vision/ # Architecture, community skills, vision
│ ├── planning/ # Implementation plans
│ └── runbooks/ # How-to guides
├── skills/ # Custom PM-Bot skills
│ ├── sdlc/ # SDLC skills (8 skills)
│ │ ├── requirements-elicitor/ # Phase 0: Vague idea → Structured requirements
│ │ ├── story-writer/ # Feature → User stories
│ │ ├── acceptance-test-generator/ # Stories → BDD tests
│ │ ├── tdd-implementer/ # RED → GREEN → REFACTOR
│ │ ├── quality-gate-checker/ # Coverage, complexity, linting
│ │ ├── security-scanner/ # OWASP, secrets, CVEs
│ │ ├── code-reviewer/ # Automated code review
│ │ ├── phase-gate-validator/ # Phase transition validation
│ │ └── feature-planner/ # Original feature planner
│ ├── tdlc/ # Backtesting, risk analysis
│ ├── rca/ # Log analysis, debugging
│ └── devops/ # Infrastructure management
├── lobster/ # Workflow pipelines
│ └── pipelines/
│ ├── sdlc-full.lobster # Original SDLC
│ ├── sdlc-tdd-full.lobster # TDD-first SDLC (NEW)
│ ├── rca-debug.lobster
│ └── tdlc-backtest.lobster
├── config/ # Configuration templates
│ └── openclaw.json
└── scripts/ # Installation & setup scripts
├── install-skills.sh
├── setup-openclaw.sh
└── pmo.service
OpenClaw workspace: ~/.openclaw/
- Phase 0 (Optional): Conversational requirements discovery for vague feature requests - transforms ideas into structured requirements using industry patterns (5 Whys, Three Amigos, Example Mapping)
- Test-Driven Development: RED → GREEN → REFACTOR cycle enforced
- Automated Quality Gates: Coverage ≥70%, complexity ≤10, linting clean
- Security Scanning: OWASP Top 10 + LLM vulnerabilities + secrets detection
- 3 Human Approval Gates: Story validation, code review, deployment
- 8 Specialized Skills: requirements-elicitor (Phase 0), story-writer, acceptance-test-generator, tdd-implementer, quality-gate-checker, security-scanner, code-reviewer, phase-gate-validator
- Phase Validation: Automated checks before advancing between phases
- Two Entry Points:
- Clear requirements → Start at story-writer (skip Phase 0)
- Vague idea → Start with requirements-elicitor → story-writer → pipeline
- Natural language commands (no rigid syntax)
- Persistent memory across sessions
- Interactive approval buttons
- llm-council: Multiple AI models create independent plans
- Consensus-based decision making
- Bias reduction through diverse perspectives
- Lobster pipelines chain skills deterministically
- Approval gates for sensitive operations
- Resumable workflows
- 3,000+ OpenClaw community skills
- Active maintenance by community
- Extensible through custom skills
- Create skill directory:
mkdir -p /opt/pm-bot/skills/custom/my-skill- Write SKILL.md:
# My Custom Skill
## Description
What this skill does
## Usage
- "Command pattern 1"
- "Command pattern 2"
## Implementation
1. Step-by-step logic
2. What tools to use
3. Expected output format- Install:
openclaw skill add /opt/pm-bot/skills/custom/my-skill# Create pipeline
cat > lobster/pipelines/my-pipeline.lobster << 'EOF'
pipeline my-pipeline {
step1 = run_skill("skill-name", {param: $input.value})
step2 = run_skill("another-skill", {data: step1.output})
return step2.result
}
EOF
# Use in Discord
"Run my-pipeline with value 'test'"openclaw status --all
tail -f ~/.openclaw/logs/gateway.log- Check Discord user ID in config
- Verify execApprovals.enabled = true
- Restart:
openclaw gateway restart
openclaw skill list
openclaw skill refresh- ✅ OpenClaw integration
- ✅ Lobster pipelines
- ✅ 6 core community skills
- ⏳ 4 custom PM-Bot skills (placeholders created)
- ✅ Discord channel
- Slack integration
- Multi-repo support
- Custom dashboards
- Fully implemented custom skills
- Web UI
- REST API
- Mobile app
- Open-source community release
- OpenClaw: https://openclaw.ai
- Lobster: https://github.com/openclaw/lobster
- Community Skills: https://github.com/VoltAgent/awesome-openclaw-skills
- PM-Bot GitHub: (TBD)
MIT License (TBD)
Built with:
- OpenClaw framework
- Lobster workflow orchestration
- Community skills (llm-council, pr-commit-workflow, tdd-guide, senior-devops, etc.)
- Claude Sonnet 4.5 (planning & implementation)
Migrated from NEXUS: 2026-02-07