rad-orchestration

Orchestrates document-driven project pipelines by signaling events and dispatching subagents.

Updated May 24, 2026
One-click install
npx skills add https://github.com/MetalHexx/rad-orc-marketplace --skill rad-orchestration-metalhexx
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: rad-orchestration
Source: https://github.com/MetalHexx/rad-orc-marketplace/tree/main/claude-plugin/skills/rad-orchestration
Command: npx skills add https://github.com/MetalHexx/rad-orc-marketplace --skill rad-orchestration-metalhexx

SYSTEM DOCUMENTATION & REQUIREMENTS

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

What problem does it solve? Running a multi-phase software project with specialized agents (coders, reviewers) requires coordinating state, events, gates, and corrective cycles. This Skill turns a main agent into a disciplined orchestrator that drives the rad-orchestration pipeline without losing track of project flow. ## Core Features & Use Cases - Event-Driven Pipeline Control: Runs the radorch.mjs CLI engine, parses JSON envelopes, executes prompted actions, and signals completion events without ever composing CLI calls manually. - Subagent Dispatch: Spawns the right coder tier (coder-junior, coder, coder-senior) and reviewer tier based on task complexity and corrective escalation rules. - Corrective Cycle Routing: Relays review report paths and handoff docs on changes_requested verdicts while the engine manages retry budgets and halts. - Use Case: After planning a project with /rad-plan, invoke /rad-execute and this Skill guides the orchestrator through every phase gate, code review, and corrective loop until final approval. ## Quick Start Ask the agent to run the rad-orchestration pipeline for a project in ~/.radorc/projects/ using the /rad-execute command.

Frequently Asked Questions about rad-orchestration

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

FAQPage Schema
How do I run a rad-orchestration project pipeline?▼

Invoke the /rad-execute skill to start, resume, or recover a run. It classifies where the operator stands and returns the commands to run, after which the orchestrator follows the event loop driven by the radorch.mjs script.

How does the orchestrator choose which coder subagent to spawn?▼

Coder tier is selected from the task's complexity field on the envelope: simple maps to coder-junior, standard and complex map to coder. coder-senior is break-glass only, reached through corrective escalation as the retry budget tightens.

What happens when a code review returns changes_requested?▼

The orchestrator signals the completion event unchanged and the pipeline engine births the corrective itself. The orchestrator only relays the review_report_path and unchanged handoff_doc into the re-spawned coder's prompt, never reading or judging findings.

Can the orchestrator edit the pipeline state.json file directly?▼

No. The state file is owned exclusively by the engine, and the CLI is the only way to observe or advance state. Even in exceptional read cases, the orchestrator must never write to it or repair values, and instead halts with a diagnosis for the operator.

What should I do when the pipeline returns an error envelope?▼

Classify the error into one of four categories using the pipeline guide, log it with the rad-log-error skill into the project's error log, then follow the category action. Category 3 and 4 errors require an immediate halt with the error shown to the operator.