Workspace orchestration for AI coding agents. Manage isolated workspaces for parallel development with git worktrees and tracked context.
- Project Initialization - Create new projects with templates, documentation, and Python tooling
- Parallel Development - Work on multiple features simultaneously without branch switching
- Isolated Environments - Each workspace has its own Python venv and dependencies
- Project Templates - Generate documentation optimized for AI agents (CLAUDE.md, ADRs, architecture docs)
- Workspace Tracking - Purpose, metadata, and timestamps per workspace
- Python 3.13
- Git
- uv (Python package manager)
git clone https://github.com/ckrough/agentspaces.git
cd agentspaces
uv sync --all-extras
uv run agentspaces --versioncd ~/projects/my-app
agentspaces workspace create main --purpose "Add user authentication"
# Created: eager-turing at ~/.agentspaces/my-app/eager-turingmkdir my-project && cd my-project
agentspaces project create -n "My Project" -d "Description"
# With Python tooling (pyproject.toml, ruff, mypy, pytest, GitHub Actions)
agentspaces project create --python -n "My CLI" -d "A CLI tool"agentspaces workspace create [branch] # Create from branch (default: HEAD)
agentspaces workspace list # List all workspaces
agentspaces workspace status <name> # Show detailed status
agentspaces workspace remove <name> # Remove workspaceagentspaces docs list # List available templates
agentspaces docs info <template> # Show template details
agentspaces docs create <template> # Generate from templateAvailable templates: readme, claude-md, architecture, development-standards, adr-template
The complete workflow from issue to PR, showing what you do and what the agent does.
Agent: Checks ready queue and claims work
bd ready --json # Check available work
bd update <issue-id> --status in_progress # Claim issueOr creates new issues with bd create based on your request
You: Run /feature-dev with your change description
Example: /feature-dev Add email validation to user registration
Agent:
- Uses
@code-explorerto find relevant code - Asks clarifying questions about approach
- Proposes implementation plan
- Decides if ADR needed (technology choices affecting data storage/processing, significant dependencies, schema changes)
- Runs
@architect-reviewerto validate design - Runs
@python-pro(for Python projects) to review architecture - Creates tasks with
bd createfor each implementation step - Gets your approval before coding
You: Approve plan, provide feedback on code
Agent:
- Updates issue status:
bd update <id> --status in_progress - Writes code and tests (TDD for clear requirements, test-after for exploration)
- If bugs discovered: creates linked issues with
bd create+bd dep add --type discovered-from - If blocker found: switches priorities immediately
- Runs tests:
uv run pytest
Coverage target: 80% on business logic, regression tests for bugs
Agent:
- Automatically runs
@simplifierto review for unnecessary complexity, duplication, naming issues - Refactors based on feedback before quality gates
You: Request commit or the agent runs automatically
Agent: Runs all checks in sequence:
uv run ruff format src/ tests/ # Format
uv run ruff check src/ tests/ # Lint
uv run bandit -r src/ # Security scan
uv run mypy src/ # Type check
uv run pytest # TestsYou: Approve changes after review
Agent:
- Runs
@code-reviewer(always) - checks bugs, security, performance - Runs
@python-pro(if Python) - validates Python idioms and patterns - Runs
@verify-app(if applicable) - E2E verification - Addresses all feedback before proceeding
- Returns to implementation if reviews fail
You: Provide final approval for commit
Agent:
- Updates docs (README.md, CLAUDE.md, ADRs)
- Exports Beads:
bd export -o .beads/issues.jsonl - Creates commit with proper prefix (bug→fix:, feature→feat:, task→task:, chore→chore:)
- Gets your explicit approval
- Commits and closes issue:
bd close <id> --reason "Implemented"
You: Run /commit-push-pr
Agent (via skill):
- Pushes commits to remote
- Creates PR with generated summary
- Returns PR URL
Quick bug fix: Skip comprehensive tests and architecture review, but must run @code-reviewer and quality gates
Prototype: Skip architecture review and comprehensive tests, but inform you what's being skipped. Security review still required.
Legacy code: Match existing patterns even if not ideal. Agent notes when deviating from standards for consistency.
Discovered blocker: Agent creates blocker issue, links with bd dep add --type blocks, and switches to blocker immediately
Workspaces are stored at ~/.agentspaces/<project>/<workspace>/:
~/.agentspaces/my-app/eager-turing/
├── .agentspace/ # Metadata
│ └── workspace.json
├── .venv/ # Isolated Python environment
└── <project files> # Git worktree
[CLI] → [Service] → [Git Worktree] → [Python Env]
- CLI Layer - Typer commands with Rich output
- Service Layer - WorkspaceService orchestration
- Infrastructure - Git and uv subprocess wrappers
See docs/design/architecture.md for detailed system design.
uv run pytest # Run tests
uv run ruff check src/ tests/ # Lint
uv run ruff format src/ tests/ # Format
uv run mypy src/ # Type checkSee CONTRIBUTING.md for development setup and guidelines.
MIT