One-click install
npx skills add https://github.com/myspecs/claude-plugins --skill dispatch-myspecs
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: dispatch
Source: https://github.com/myspecs/claude-plugins/tree/main/plugins/myspec-factory/skills/dispatch
Command: npx skills add https://github.com/myspecs/claude-plugins --skill dispatch-myspecs

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) and references (resource) components.

What problem does it solve? Coordinating parallel AI coding workers on a spec bundle is error-prone: tasks that share files conflict across branches, workers start with zero context, and sessions get lost after launch. This Skill groups related tasks into workers, builds each worker a self-contained brief from the spec, dispatches cloud or local sessions, and records every session for later steering. ## Core Features & Use Cases - Task grouping: Clusters ready tasks by dependency chains, shared files or modules, and requirement ids so one worker, one branch, and one pull request cover related work without conflicts. - Brief assembly: Uses build-brief.py to copy task blocks, requirement texts, and owner decisions verbatim from the local spec bundle into a worker brief under roughly 400 lines. - Multi-path dispatch: Starts workers as remote cloud sessions (Agent tool or claude --cloud) with local git worktree sessions as fallback, respecting the concurrency cap and run policy. - Session bookkeeping: Records every session in factory-sessions.json via registry.py, arms a pull-request monitor per worker, and hands ready pull requests to the integrate skill. - Use Case: After the plan skill produces a wave for milestone 2, say "dispatch the wave" and the manager groups tasks 45-47 into one worker, writes its brief, starts a cloud session, and posts the board with session URLs. ## Quick Start Ask the AI to dispatch the approved wave of tasks from the plan skill, for example by saying "dispatch the wave for bundle checkout".

Frequently Asked Questions about dispatch

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I dispatch parallel Claude Code worker sessions for spec tasks?▼

Group related ready tasks by dependencies, shared files, and requirement ids, then start one cloud session per group with claude --cloud or a remote Agent call. Each worker gets a self-contained brief built from the spec bundle, and every session is recorded in factory-sessions.json.

How are tasks grouped into workers in a software factory?▼

Tasks are grouped when they form a dependency chain, touch the same module or files, or implement the same requirement ids. Groups never cross a milestone or a declared pull-request split, and a single-task worker is used only when nothing relates to it.

When should I use local worktree workers instead of cloud sessions?▼

Use local worktree sessions only when the run policy disallows cloud dispatch, the user requests it, or the cloud refuses for a lasting reason such as missing claude.ai sign-in. Local workers share the machine's CPU and memory, so consider lowering the concurrency cap.

Why does claude --cloud fail in a background shell?▼

claude --cloud requires an interactive terminal and exits 1 from a plain background Bash call. Wrap it with script -q to provide a pseudo-terminal, and always cd into a local clone of the target repository in the same command.

What happens when a dispatched worker reports BLOCKED or fails?▼

A BLOCKED: spec report pauses dispatch while the manager asks the owner, records the clarification in tasks.md, and redispatches. A failed worker is redispatched once with the failure output added to the brief; a second failure stops the lane and is reported.