Skip to main content

​
Workspace Commands

Create and manage isolated workspaces for agent work.

​
Overview

Workspaces are git worktrees - isolated working directories for each issue/feature. Each workspace:
  • Has its own feature branch (e.g., feature/min-123-add-login)
  • Shares git history with the main repo (no separate clone)
  • Can run independently (separate node_modules, builds, etc.)
  • Is located at {project}/workspaces/{issue-id}/
This allows multiple agents to work on different features simultaneously without conflicts.

​
pan workspace create

Create a new workspace (git worktree) without starting an agent: 
pan workspace create MIN-123

# Create workspace and start Docker containers
pan workspace create MIN-123 --docker

​
Docker Integration

The --docker flag automatically starts containers after workspace creation: 
pan workspace create MIN-123 --docker
What it does:
  1. Creates the workspace (git worktree or custom command)
  2. Looks for docker-compose files in:
    • {workspace}/docker-compose.yml
    • {workspace}/docker-compose.yaml
    • {workspace}/.devcontainer/docker-compose.yml
    • {workspace}/.devcontainer/docker-compose.devcontainer.yml
    • {workspace}/.devcontainer/compose.yml
  3. Runs docker compose -p "{project}-feature-{issue}" -f {file} up -d --build
Docker Project Naming: Each workspace gets a unique Docker Compose project name to avoid container conflicts:
  • Format: {project-name}-feature-{issue-id} (e.g., mind-your-now-feature-min-123)
  • The project name comes from name in your ~/.panopticon/projects.yaml
  • Container names follow: {project}-{service}-1 (e.g., mind-your-now-feature-min-123-api-1)
This allows multiple workspaces to run simultaneously without port or name conflicts. Why this matters:
  • Containers start warming up while you review the issue
  • Environment is ready when the planning agent starts asking questions
  • You can test assumptions during planning without waiting for builds
Dashboard integration: The planning dialog includes a “Start Docker containers” checkbox:
  • Default: Enabled (containers start automatically)
  • Preference saved: Your choice is remembered in browser localStorage
  • Key: panopticon.planning.startDocker
To change the default via browser console: 
// Disable Docker by default
localStorage.setItem('panopticon.planning.startDocker', 'false');

// Enable Docker by default (this is the out-of-box default)
localStorage.setItem('panopticon.planning.startDocker', 'true');
Example workflow: 
# From dashboard: click "Start Planning" (Docker enabled by default)
# Or from CLI:
pan workspace create MIN-123 --docker

# While containers build in background:
# - Review the Linear issue
# - Check related PRs
# - Think about approach

# By the time you're ready to engage with the planning agent,
# the dev environment is warm and ready for testing

​
pan workspace list

List all workspaces and their status: 
pan workspace list

​
pan workspace destroy

Clean up and remove a workspace: 
pan workspace destroy MIN-123

# Force destroy (even with uncommitted changes)
pan workspace destroy MIN-123 --force

​
Main Project Branch Convention

IMPORTANT: The main project directory should ALWAYS stay on the main branch. All feature work happens in workspaces (git worktrees), never in the main project.
  • Main project: Always on main - serves as the stable reference point
  • Workspaces: Each has its own feature branch checked out
  • Want to check something? Create a worktree, don’t checkout in main
  • Quick test? Create a worktree
  • Emergency hotfix? Create a worktree
This convention ensures:
  1. Specialists (review-agent, test-agent) always find the correct code
  2. Merge operations have a clean target
  3. Multiple agents don’t interfere with each other
If you see a warning like “Main project is on branch ‘feature/xxx’ instead of ‘main’”, something has incorrectly checked out a feature branch in the main project. Fix it with git checkout main.

​
Git Hooks for Branch Protection

Panopticon provides git hooks to enforce the main branch convention. These hooks are automatically installed:
  • On pan project add - Hooks are installed when registering a new project
  • On pan sync - Hooks are updated in all registered projects
The hooks are polyrepo-aware and will install in all git repositories within your project. Auto-revert behavior:
  • Agents & planning sessions: Automatically revert to main if they accidentally checkout a feature branch
  • Human users: Display a prominent warning (no auto-revert by default)
To enable auto-revert for humans: 
export PANOPTICON_AUTO_REVERT_CHECKOUT=1  # Add to .bashrc/.zshrc
Manual hook installation (for projects not in registry): 
./scripts/install-git-hooks.sh /path/to/your/project

​
Workspace Structure

Workspaces include planning artifacts stored inside the workspace directory: 
workspaces/feature-min-123/
├── .planning/                 # ⚠️ NOT tracked in git (ephemeral)
│   ├── STATE.md               # Current planning state
│   ├── output.jsonl           # Full conversation history
│   ├── PLANNING_PROMPT.md     # Initial planning prompt
│   ├── CONTINUATION_PROMPT.md # Context for continued sessions
│   └── output-*.jsonl         # Backup of previous rounds
└── ... (code)
⚠️ Note: .planning/ is currently ephemeral (not tracked in git) The .planning/ directory is listed in .gitignore to prevent cross-contamination between issues. When branches merge, planning state from issue A could contaminate issue B’s workspace, causing agents to work on the wrong tasks. Current behavior: Planning state is local to your machine only. If you switch machines, you’ll need to re-run the planning session. See #111 for the future enhancement to support cross-machine planning sync safely.
When the planning session completes, the AI generates:
  • STATE.md - Current understanding, decisions made, and implementation plan
  • Beads tasks - Trackable sub-tasks with dependencies for the implementation
  • Feature PRD - Copied to docs/prds/active/{issue}-plan.md for documentation
What works:
  1. Context recovery - If Claude’s context compacts, STATE.md and beads preserve the planning decisions
  2. Audit trail - See decisions made in STATE.md
  3. Beads tasks - Sub-tasks created during planning ARE persisted (beads uses label-based isolation)

​
Safeguards

​
Planning vs Work Agent Safeguard

The dashboard prevents starting a work agent (agent-<issue>) while a planning agent (planning-<issue>) is still running for the same issue. This prevents conflicts where both agents try to modify the same workspace. If you see “Planning agent is still running” when trying to start work:
  1. Complete the planning session - Use “Complete Planning” button
  2. Or abort planning - Use “Abort Planning” if you want to discard
  3. Or kill the session manually - tmux kill-session -t planning-<issue>

​
Specialist Agent Safeguards

All specialist agents (review-agent, test-agent, merge-agent) are configured to:
  1. Run in workspaces only - Never in the main project directory
  2. Never checkout branches - They work with the branch already in the workspace
  3. Create workspaces if needed - Use pan workspace create <ISSUE-ID> instead of git checkout
These safeguards are enforced through:
  • Prompt instructions - Clear warnings in all specialist prompts
  • Code enforcement - Specialists spawn with workspace as cwd
  • Git hooks - Auto-revert if checkout detected in tmux sessions