skills.rest
akoita
akoitaCOMMUNITY
Explore all skills by this author →
𝕏@aboobakarGitHub

maestro

Orchestrates development work in Claude Code by delegating bounded tasks to subagents.

Updated Jul 9, 2026
Search Tags:#workflow#code review#claude code#subagents#delegation#orchestration#model selection
One-click install
npx skills add https://github.com/akoita/agent-toolkit --skill maestro-akoita
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: maestro
Source: https://github.com/akoita/agent-toolkit/tree/main/plugins/claude/maestro/skills/maestro
Command: npx skills add https://github.com/akoita/agent-toolkit --skill maestro-akoita

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Non-trivial development work in Claude Code often overloads a single session with implementation details, leaving too little context for judgment, review, and verification. This Skill keeps requirements, architecture, planning, and publication in the main session while routing bounded implementation and exploration to reusable subagents. ## Core Features & Use Cases - Model and effort routing: Selects between Fable, Opus, Sonnet, and Haiku aliases with documented escalation triggers so expensive tiers are used only when justified. - Orchestration primitives: Chooses between direct main-session work, Agent-tool subagents, experimental agent teams, and dynamic workflows based on task shape. - Five-phase workflow: Covers analysis and planning, bounded delegation with self-contained worker contracts, main-session review of every diff, targeted fix rounds, and verified publication. - Use Case: When implementing a multi-file feature, the main session writes a file-level plan, delegates two independent work items to Opus subagents, reviews their diffs, runs verification commands itself, and only then commits and reports. ## Quick Start Ask the agent to orchestrate a non-trivial feature or refactor using the maestro workflow, delegating bounded implementation items to subagents while keeping review and commits in the main session.

Frequently Asked Questions about maestro

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

FAQPage Schema
How do I orchestrate subagents in Claude Code for a large feature?▼

Write a file-level plan in the main session, split it into independent work items, and delegate each to a subagent with a self-contained contract covering task, context, constraints, and verification commands. Review every returned diff yourself before accepting it.

Which Claude model should I use for orchestration versus implementation?▼

Use Opus at high effort as the default for the main session and correctness-sensitive workers, Sonnet for mechanical low-risk changes, and Haiku for read-only exploration. Escalate to Fable only for strategic, ambiguous, or high-stakes decisions where a wrong framing is costly.

When should I use agent teams instead of subagents in Claude Code?▼

Use experimental agent teams only when workers must share findings, challenge hypotheses, or coordinate across layers, and only after user approval with CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1. For sequential work or simple report-back flows, plain subagents are cheaper and sufficient.

When should I not delegate work to subagents?▼

Skip delegation for trivial localized edits, pure analysis or review with no implementation, strictly sequential tasks, explicit user requests for direct implementation, and small corrections found during review. Delegation adds overhead without benefit in these cases.

What are the limitations of subagent worktree isolation?▼

A subagent worktree branches from the repository's default branch rather than the parent's current HEAD unless worktree.baseRef is set to head, and untracked or gitignored files are absent. The maestro must explicitly inspect and integrate worktree changes.

Related Skills

Explore More Software Engineering Skills→
Pyl-TechPyl-Tech

spec-gate

Verify implementation specs are complete and actionable before AI coding begins.

Official
Advanced
real-casereal-case

task-deliver

Commits verified changes and opens a pull request after a tool-backed verification gate passes.

Community
Advanced
VanLVanL
1

cso

Automate security audits and threat verification across infrastructure, code, and AI systems.

Community
Advanced
nnexainnexai
1

gsd-fast

Executes minor refactors and simple fixes directly without overhead.

Community
Intermediate