Table of Contents

How to Develop in Parallel with Multiple Agents

Parallel development lets you move independent work packages (WPs) at the same time while keeping each workspace isolated. Spec Kitty's worktree-per-WP model makes this safe and predictable.

Why Parallel Development?

  • Shorten delivery time by running independent WPs concurrently.
  • Keep changes isolated to avoid accidental cross-contamination.
  • Use the dashboard to coordinate and rebalance work in real time.

Prerequisites

  • A feature with multiple WPs in lane: "planned".
  • Multiple terminals or agents available.
  • Dependencies defined in WP frontmatter.

Identifying Parallel Opportunities

  1. List WPs and their dependencies.
  2. Start WPs that do not depend on each other.
  3. Hold any WP that depends on unfinished work.

Example: Two Independent WPs

Terminal 1 - Agent A

spec-kitty agent workflow implement WP01
cd .worktrees/###-feature-WP01
# Agent A implements WP01

Terminal 2 - Agent B (simultaneously)

spec-kitty agent workflow implement WP02
cd .worktrees/###-feature-WP02
# Agent B implements WP02

Example: Fan-Out Pattern

        WP01
      /  |  \
   WP02 WP03 WP04

Once WP01 is finished, three agents can work on WP02, WP03, and WP04 in parallel.

Example: Dependent WPs

# Agent A completes WP01 first
spec-kitty agent workflow implement WP01
# ... implement and finish WP01

# Agent B starts WP02 after WP01 exists
spec-kitty implement WP02 --base WP01
cd .worktrees/###-feature-WP02

Best Practices

  • Start with dependency-free WPs, then fan out.
  • Communicate when base WPs complete so dependents can start.
  • Keep each agent in its own worktree path.
  • Use workflow commands to keep lane history and dashboard accurate.

Monitoring Parallel Work

In your terminal:

spec-kitty agent tasks status

Or in your agent:

/spec-kitty.status

Use the dashboard to monitor lane movement and agent activity in real time.

Command Reference

See Also

Background

Getting Started

Edit this page