ClaudeAutoPM
AI-Powered Project Management Framework for Claude Code
ClaudeAutoPM transforms your development workflow with intelligent automation, specialized AI agents via a plugin system, and complete GitHub/Azure DevOps integration. Install only what you need — pay only for the tokens you use.
Quick Install
Requires Node.js >= 22 and npm >= 10.
npm install -g claude-autopm
cd your-project
autopm install
Choose your scenario during installation:
| Scenario | Agents | Use case |
|---|
| Lite | 7 core | Local PM only, minimal token footprint |
| GitHub | 12 | Core + languages for GitHub projects |
| Azure | 12 | Core + languages + Azure DevOps sync |
| Docker | 26 | Core + languages + frameworks + testing + DevOps |
| Obsidian | 7 core | PM + Obsidian vault integration |
| Full | 45+ | Complete CI/CD: cloud, databases, AI included |
| Performance | 55+ | All plugins (incl. data + ML) + max parallelization |
Architecture
Plugin-Based Agent System
ClaudeAutoPM uses a plugin architecture for token efficiency. Only 7 core agents are loaded by default. Specialized agents are installed on demand via plugins.
autopm/.claude/agents/ ← 7 core agents (always available)
packages/plugin-*/agents/ ← 50+ specialized agents (per plugin)
Token savings: Lite installs use 73% fewer agent tokens vs full.
XML Rules Engine
8 critical rules auto-loaded via @include in XML format:
| Rule | Enforcement |
|---|
tdd.enforcement.xml | Zero-tolerance TDD cycle |
coverage-thresholds.xml | 80/75/80/80 coverage gates |
agent-mandatory.xml | Agent delegation + context strategy |
context7.xml | Context7 MCP before any library code |
github-operations.xml | Repo protection + mandatory issue footer |
naming-conventions.xml | Naming prohibitions + code quality |
command-pipelines.xml | 5 mandatory command sequences |
issue-structure.xml | Standard GitHub issue structure (goal, criteria, verification) |
3 MD operational guides: standard-patterns.md, frontmatter-operations.md, git-strategy.md
Dynamic Agent Registry
agent-registry.xml is generated at install time based on selected plugins:
<!-- Core agents (always present) -->
<agents category="core">
<agent name="code-analyzer">...</agent>
<agent name="test-runner">...</agent>
...
</agents>
<!-- Plugin agents (injected by installer) -->
<!-- PLUGIN_AGENTS_START -->
<agents category="languages">...</agents>
<agents category="frameworks">...</agents>
<!-- PLUGIN_AGENTS_END -->
Core Agents (Always Available)
| Agent | Purpose |
|---|
agent-manager | Agent lifecycle, registry maintenance |
code-analyzer | Code analysis, bug detection, logic tracing |
test-runner | Test execution with actionable insights |
file-analyzer | Summarize large files, reduce context |
parallel-worker | Multi-agent parallel work streams |
mcp-manager | MCP server install, config, health |
context-optimizer | Context window efficiency, compaction |
Plugin Agents
| Plugin | Agents | Examples |
|---|
| plugin-languages | 5 | python-backend-engineer, nodejs-backend-engineer, bash-scripting-expert |
| plugin-frameworks | 6 | react-frontend-engineer, tailwindcss-expert, e2e-test-engineer |
| plugin-cloud | 8 | aws-cloud-architect, kubernetes-orchestrator, terraform-infrastructure-expert |
| plugin-devops | 7 | docker-containerization-expert, github-operations-specialist, observability-engineer |
| plugin-databases | 5 | postgresql-expert, mongodb-expert, redis-expert |
| plugin-ai | 8 | openai-python-expert, gemini-api-expert, langchain-expert |
| plugin-data | 3 | airflow-orchestration-expert, kedro-pipeline-expert |
| plugin-testing | 1 | frontend-testing-engineer |
| plugin-ml | 10 | pytorch-expert, tensorflow-keras-expert, scikit-learn-expert |
| plugin-pm | — | 29 PM commands (PRD, epic, issue management) |
Quick Start
All commands run inside Claude Code (not terminal):
# Initialize context and testing
/context:create
/test:test-setup
# PM workflow
/pm:issue-start 42
/pm:issue-close 42
# Working with agents
@python-backend-engineer Build FastAPI endpoint
@test-runner execute all tests
@code-analyzer review recent changes
Features
14 Official Plugins
- Core — Context management, testing, configuration, MCP
- PM System — PRD, epics, issues, decomposition, git workflow
- Obsidian — Read-only vault sync, Dataview frontmatter, diagnostics
- Azure DevOps — Work items, boards, pipelines, sprints
- GitHub — Workflows, issues, PRs, automation
- Languages — Python, JavaScript, Node.js, Bash
- Frameworks — React, Tailwind, UX, E2E testing, NATS
- Testing — Frontend unit/integration, E2E
- DevOps — Docker, Kubernetes, Traefik, SSH, observability
- Cloud — AWS, Azure, GCP, Terraform, serverless
- Databases — PostgreSQL, MongoDB, BigQuery, CosmosDB, Redis
- AI — OpenAI, Gemini, Claude, LangChain, HuggingFace
- ML — PyTorch, TensorFlow/Keras, scikit-learn, AutoML
- Data — Airflow, Kedro, LangGraph
Provider Integration
- GitHub — Issues, PRs, Actions, Projects
- Azure DevOps — Work Items, Boards, Pipelines
- Local — Git-based workflow with markdown tracking
Execution Strategies
- Sequential — Safe, predictable
- Adaptive — Intelligent mode selection (default)
- Hybrid — Maximum parallelization
Documentation
Why ClaudeAutoPM?
| Feature | ClaudeAutoPM | Traditional Tools |
|---|
| AI-native | Built for Claude Code | Adapted/retrofitted |
| Token-efficient | Plugin system, install what you need | Monolithic |
| Agents | 50+ specialized, load on demand | Generic or none |
| Rules | 8 XML auto-enforced (TDD, coverage, Context7) | Manual checklists |
| Integration | GitHub + Azure DevOps | Limited |
| Documentation | Context7-verified, always current | Often outdated |
Contributing
We welcome contributions! See Contributing Guide.
TDD is mandatory — every PR must follow Red-Green-Refactor.
License
MIT License - see LICENSE for details.
Links
