User Guide Overview
This guide covers Warpai’s primary workflows across software engineering and multi-agent cross-domain teams: from concurrent feature development in isolated Git worktrees and omnidirectional Project mode to integrated terminal editing, peer review pipelines, and remote SSH/WSL setups.
Workflow 1: Parallel Development with Git Worktrees (For Software Coding)
Warpai’s flagship workflow for software repositories is Worktree Mode. By allocating each feature to an isolated physical Git worktree, multiple CLI agents write code concurrently without dirtying branches or invalidating build caches.
┌─────────────────────────────────────────────────────────────────────────────┐
│ │
│ 1. Switch to "Worktree" mode at the top of the collaboration panel │
│ 2. Create isolated physical worktrees for each feature (api, ui, tests) │
│ 3. Launch your preferred CLI agent inside each corresponding tab │
│ 4. Select a Coordinator agent from the dropdown │
│ 5. Coordinator scopes requirements and dispatches tasks to worktrees │
│ 6. Workers develop concurrently in isolation and submit diffs for review │
│ 7. Coordinator verifies results and integrates changes into the main repo │
│ │
└─────────────────────────────────────────────────────────────────────────────┘
Step-by-Step Instructions:
1. Create Physical Worktrees for Features
Run standard Git commands or use Warpai’s worktree wizard:
git worktree add ../my-project-api feature/rest-api
git worktree add ../my-project-ui feature/frontend-components
git worktree add ../my-project-tests feature/integration-tests
Each worktree opens in its own vertical tab, locking its working directory.
2. Launch Agents in Each Worktree
Switch to each tab and launch your chosen agent:
- In
../my-project-api: launchclaudeorcodex - In
../my-project-ui: launchcursor-agentortrae - In
../my-project-tests: launchgeminioragy
3. Designate the Coordinator
In the left collaboration panel:
- Switch the mode toggle from Project to Worktree.
- Click the Coordinator dropdown and select an agent with strong architectural oversight.
- The Coordinator receives a unified view of all active worktree branches and agent states.
4. Dispatch Requirements and Supervise Execution
Provide your high-level objective to the Coordinator:
“Implement a data export feature. Delegate the REST API to Worker A, frontend download components to Worker B, and integration tests to Worker C. Have them peer-review each other before final integration.”
The Coordinator dispatches tasks via Warpai’s native MCP tool (warp_task_assign). Workers automatically pull tasks when idle.
Workflow 2: Omnidirectional Project Mode (For Cross-Disciplinary & Non-Coding Teams)
For non-coding and multi-agent research domains (scientific paper drafting, creative design, quantitative finance, patent drafting, content operations, and business intelligence), rigid Git branching is unnecessary. Warpai delivers full first-class support for Project Mode:
┌─────────────────────────────────────────────────────────────────────────────┐
│ │
│ 1. Switch to "Project" mode at the top of the collaboration panel │
│ 2. No Git repo required: launch agents inside any folder or cloud drive │
│ 3. Break rigid hierarchies: all agents act as Commanders or Workers │
│ 4. Direct two-way messaging across any agent (warp_agent_send / reply) │
│ 5. Structured Q&A channels for peer inquiries (warp_agent_ask / answer) │
│ 6. Autonomous task claiming from public pools (warp_task_create_pool) │
│ 7. Expose verified milestones via shared memory (warp_context_publish) │
│ │
└─────────────────────────────────────────────────────────────────────────────┘
- Omnidirectional Messaging & Tasks: Every agent operates as an equal peer node. In scientific research, a literature retrieval agent supplies citations directly to the manuscript author and recommends plot aesthetics to the data visualization agent, while a mock peer reviewer dispatches revision notices.
- Dynamic Role Reversals: An agent can coordinate a roadmap as Commander in stage 1, then transition to Worker in stage 2 under a specialized QA lead.
- Dedicated Implementation Manuals: For step-by-step procedures and templates, see our dedicated guides on Project Mode: Omnidirectional Collaboration & Cross-Domain Guide and Worktree Mode: Parallel Git Software Engineering.
Workflow 3: Automated Peer Review & Integration Pipeline
Inside Worktree Mode, completed tasks are never blindly merged into main. Instead, Warpai enforces an automated peer review loop:
- Submission: When a worker finishes coding, it calls
warp_task_submitwith test logs (warp_evidence_add) and git diffs. - Review Assignment: Coordinator assigns another available agent as reviewer (
warp_task_review). - Review Verdict:
needs_changes: Reviewer flags issues or anti-patterns; worker receives actionable feedback and refactors.accepted: Reviewer approves the changes.
- Integration: Coordinator calls
warp_task_integrateto merge the branch intomain, execute regression tests, and create a clean Git commit.
Workflow 4: The Terminal-Integrated Workbench
Keep your hands on the keyboard without switching windows:
- Project Explorer: Browse, create, and manage files alongside your terminal panes.
- Integrated Code Editor & Vim Modes: Edit configuration files and scripts directly with
:wand:qkeybindings. - Live Markdown Preview: Render technical documentation, tables, and diagrams instantly.
- Native Git Diff Inspector: Review modifications side-by-side before committing.