manager

Decomposes architecture plans into granular task files for worker agents.

Updated May 2, 2026
One-click install
npx skills add https://github.com/whinchman/midi-man-mk3 --skill manager-whinchman
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: manager
Source: https://github.com/whinchman/midi-man-mk3/tree/main/.claude/skills/manager
Command: npx skills add https://github.com/whinchman/midi-man-mk3 --skill manager-whinchman

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Turning a high-level architecture plan into concrete, assignable work is slow and error-prone. This Skill breaks plans and design specs into small, independently testable task files that Coder, QA, Designer, and Automation agents can execute in parallel without cross-repo confusion. ## Core Features & Use Cases - Plan Decomposition: Reads architecture plans from .workflow/plans/ and produces structured task files with type, status, repo, parallel group, branch naming, and acceptance criteria. - Parallel Group Assignment: Groups tasks by dependency order so the Coordinator can spawn independent tasks simultaneously across repos and mono-repo sub-projects. - Contract Embedding: Extracts cross-repo API shapes, data models, and event schemas and embeds them verbatim into each task file so execution agents never need to read other repos. - Dual Workflow Backends: Supports a default markdown backend writing to .workflow/tasks/ and a github_project mode that delegates task creation to a board-man agent as GitHub sub-issues. - Use Case: Given a multi-repo feature plan touching a backend service and a mobile app, generate task files where schema changes land in group 1, backend work in group 2, and mobile UI in group 3, each with its own feature sub-branch. ## Quick Start Ask the agent to read the feature plan in .workflow/plans/ and decompose it into parallel-grouped task files for the affected repos.

Frequently Asked Questions about manager

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

FAQPage Schema
How do I break an architecture plan into tasks for AI agents?▼

Read the plan from .workflow/plans/ along with agent.yaml, identify affected repos and cross-repo contracts, then write one task file per unit of work in .workflow/tasks/. Each task gets a type, parallel group, branch name, acceptance criteria, and embedded interface contracts.

What is a parallel group in multi-agent task decomposition?▼

A parallel group is an integer assigned to each task indicating execution order. Tasks in the same group have no dependencies and run simultaneously; group 2 starts only after all group 1 tasks complete. Schema changes typically go in group 1, backend in group 2, frontend in group 3.

Can task files be created as GitHub issues instead of markdown?▼

Yes. Setting workflow.backend to github_project in agent.yaml switches storage to GitHub sub-issues. The manager delegates to a board-man agent, which creates sub-issues with the same markdown body schema, links them via GraphQL addSubIssue, and sets status fields.

How are mono-repo sub-projects handled in task routing?▼

Mono-repo parents declare a projects list in their agent-build.yaml, and each sub-project is addressed with dotted notation like mobile.ios or mobile.android in the task's Repo field. Sub-project tasks are independent and can share the same parallel group.

What are the limitations of the manager agent?▼

The manager never writes application, test, or infrastructure code and never merges to the default branch or pushes to remote. It only creates task files, commits them on a feature branch, and reports ambiguities back for human judgment.