orchestrate

Orchestrates GitHub issue-driven sessions with worker dispatch, adversarial review, and labeled PR shipping.

Updated Aug 1, 2026
One-click install
npx skills add https://github.com/tonytino/aubreyshungry --skill orchestrate-tonytino
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: orchestrate
Source: https://github.com/tonytino/aubreyshungry/tree/main/.claude/skills/orchestrate
Command: npx skills add https://github.com/tonytino/aubreyshungry --skill orchestrate-tonytino

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Coordinating multi-step repository work across issues, subagents, reviews, and pull requests is error-prone without a defined process. This Skill enforces a consistent session lifecycle so every task is tracked, reviewed, and shipped according to repo governance rules. ## Core Features & Use Cases - Issue Recon and Claiming: Searches existing GitHub Issues first to prevent duplicates, then claims and branches from the issue. - Worker Dispatch by Model Tier: Decomposes tasks into specced work units and assigns subagents to model tiers, reserving the most capable tier for food content. - Adversarial Review and Safe Shipping: Runs a fresh reviewer each round with a 2-round cap, then ships a Conventional-Commit PR labeled safe:agent or safe:human with changelog and owner-gate checks. - Use Case: Starting a session to implement a new meal-plan feature: the Skill finds or creates the issue, dispatches workers, reviews their output adversarially, and merges the PR autonomously or hands it off to a human when owner-gated surfaces are touched. ## Quick Start Start the session by running the orchestrate workflow to triage open GitHub issues and dispatch workers for the top-priority task.

Frequently Asked Questions about orchestrate

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

FAQPage Schema
How do I orchestrate multi-step coding tasks with AI subagents?

Decompose the task into work units with explicit specs and acceptance criteria, then dispatch each unit to a subagent at an appropriate model tier. Review every worker output with a fresh adversarial reviewer before shipping the changes as a labeled pull request.

How to manage GitHub issues in an agent-driven workflow?

Search existing issues first with gh issue list to avoid duplicates, then claim the issue by assigning yourself and relabeling it status:in-progress. Branch as issue-<n>-<slug> and close the issue automatically with Closes #<n> in the PR.

When can the orchestration loop be skipped?

The loop can be skipped only for the tiny-task exception: answering questions and typo-class or one-line documentation fixes. Committed changes still ship per PR conventions using the sanctioned skip-review bypass.

What is the difference between safe:agent and safe:human PR labels?

safe:agent PRs are babysat through CI and squash-merged autonomously on green. safe:human PRs are driven to green CI and then handed to a human, and are mandatory whenever the diff touches owner-gated surfaces.

Why must a reviewer never be the same agent as the worker?

A fresh adversarial reviewer each round prevents self-approval bias and catches defects the worker is blind to. The review loop is capped at two rounds, with unresolved items escalated in the PR description.