dispatch

Orchestrates worker agents across git worktrees to build and integrate ticketed feature work.

3|Updated Jan 10, 2026
One-click install
npx skills add https://github.com/MaxWolf-01/agents --skill dispatch-maxwolf-01
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: dispatch
Source: https://github.com/MaxWolf-01/agents/tree/main/mx/skills/dispatch
Command: npx skills add https://github.com/MaxWolf-01/agents --skill dispatch-maxwolf-01

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) components.

What problem does it solve? Coordinating multiple AI coding agents on a feature's ticket backlog is error-prone: claims collide, parallel edits conflict, and finished work goes unreviewed. This Skill provides a single-orchestrator loop that claims tickets, spawns worker agents in isolated git worktrees, integrates their branches, and routes every build through a user ruling before merge. ## Core Features & Use Cases - Ticket DAG orchestration: Computes the frontier of unblocked tickets from blocked-by dependencies, plans parallel waves based on file-overlap and surface-coherence analysis, and spawns one worker per ticket on local or remote (SSH) hosts. - Review-gated integration: Workers flip tickets to review when done; the orchestrator reads the diff, presents a review page, and merges only on user accept, with amend, redo, and reject paths. - Failure recovery: Probes worker sessions, resumes stopped workers with guidance, resets unrecoverable sessions to open, and re-arms watchers after orchestrator restarts. - Use Case: A feature spec has been broken into eight tickets. Run dispatch to work them serially or in waves: each ticket gets its own worktree and worker agent, landed branches merge into the feature branch, and the feature closes with a whole-feature code review, hardening pass, and debrief. ## Quick Start Say "dispatch the tickets for this feature" after creating tickets with the tracker workflow, and the orchestrator will claim, spawn, integrate, and report until the feature ships.

Frequently Asked Questions about dispatch

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

FAQPage Schema
How do I run multiple AI coding agents in parallel on one feature?

Use the dispatch loop: it computes the frontier of unblocked tickets, plans waves by checking file overlap and surface coherence, then spawns one worker agent per ticket in its own git worktree. Each worker's branch is integrated back into the feature branch only after your review ruling.

How does dispatch prevent merge conflicts between parallel agents?

Dispatch estimates which files each frontier ticket will touch and only spawns tickets with disjoint edits together; doubtful pairs are serialized. It also treats tickets sharing one UI, document, or API surface as a coherence risk and assigns them to a single serial worker.

Can dispatch run worker agents on remote machines?

Yes. The worker-hosts command lists candidate machines with capability records, and dispatch ctl runs spawn, probe, and cleanup over SSH on the chosen host. Hosts marked isolated run workers with bypassPermissions mode; hosts without records use default permissions.

What happens when a dispatched worker agent crashes or stops early?

The orchestrator probes sessions to detect exits, reads the worker's log to diagnose the stop, and resumes the exact conversation once blockers are cleared. Unrecoverable sessions are reset to open so a fresh worker can take the ticket, since the ticket and spec carry all needed context.

When should I use serial dispatch instead of parallel waves?

Use a wave size of one for small features, surface-heavy work where tickets share a UI or API façade, or whenever parallel-safety is doubtful. Serial mode keeps the same orchestrator benefits: frontier bookkeeping, pre-merge review, integration, and status tracking.