skills.rest
williamthorsen
williamthorsenCOMMUNITY
Explore all skills by this author →
WebsiteGitHub

orchestrate

Orchestrates multi-phase development pipelines by delegating work to specialized subagents.

1|1|Updated Feb 25, 2026
Search Tags:#code review#workflow automation#subagents#delegation#mcp#orchestration#pipeline
One-click install
npx skills add https://github.com/williamthorsen/codeassembly --skill orchestrate-williamthorsen
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: orchestrate
Source: https://github.com/williamthorsen/codeassembly/tree/main/packages/agents/content/skills/orchestrate
Command: npx skills add https://github.com/williamthorsen/codeassembly --skill orchestrate-williamthorsen

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Coordinating architecture, planning, implementation, and iterative code review across a development task requires constant manual delegation and tracking. This Skill acts as a pipeline execution engine that delegates all work to specialized subagents and manages run state, artifacts, and review cycles automatically. ## Core Features & Use Cases - Pipeline execution engine: Runs configurable phases (architecture, planning, implementation, review-cycle) by dispatching specialized subagents via the Task tool and using their structured output for flow control. - Iterative review cycle: Dispatches parallel reviewers (core, code, silent-failure, test), aggregates findings by criticality, delegates fixes to a coder subagent, and loops until convergence or budget exhaustion. - Run tracking and artifacts: Initializes runs via MCP tools, writes sequenced artifacts (run-manifest, plans, reviews, change summaries), and emits lifecycle events for observability. - Use Case: Given a ticket describing a new feature, the engine analyzes the task, optionally runs architecture and planning phases, implements the code through a coder subagent, then runs parallel reviews and fix cycles until the change passes the approval threshold. ## Quick Start Invoke the orchestrate engine through a wrapper skill such as orchestrate-dev with a task description and a pipeline specification listing the phases to run.

Frequently Asked Questions about orchestrate

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

FAQPage Schema
How do I run a multi-phase development pipeline with subagents?▼

Invoke the engine through a wrapper skill like orchestrate-dev with a task description and a pipeline specification listing phases such as architecture, planning, implementation, and review-cycle. The engine delegates each phase to specialized subagents and manages flow control from their structured output.

How does the iterative code review cycle work?▼

The engine dispatches a core reviewer and aspect reviewers in parallel, aggregates findings by criticality, and delegates fixes to a coder subagent when findings meet the approval or budget threshold. It repeats fix and re-review rounds until convergence or the max-review-rounds budget is exhausted.

Can I assign different models to different subagents?▼

Yes, pass --models=key:model pairs or set orchestration.models in preferences.yaml. Keys include coder, architect, planner, reviewer, and savings_analyzer, with engine defaults such as opus for coder and haiku for savings_analyzer.

What happens when the MCP server is unavailable?▼

Behavior depends on the mcp_policy preference: required aborts the run, optional continues without tracking after a notice, and prompt asks the developer before continuing. Without MCP, no run-index.json, run-log.jsonl, or Factory visualization is produced.

How are external plans handled before implementation?▼

The engine detects external plans in the task description and evaluates provenance and codebase freshness to assign a trust tier of high, medium, or low. High-trust plans skip planning and are converted into canonical plan artifacts, while low-trust plans force architecture and planning phases to run.

What happens when a reviewer subagent runs out of turns?▼

The engine detects the interruption via a pending criticality sentinel in the review artifact and dispatches one constrained retry with a file allow-list and budget warning. If the retry also fails, the dispatch is recorded as failed and contributes medium criticality to aggregation.

Related Skills

Explore More Software Engineering Skills→
jeffutterjeffutter
3

tk-planner

Automate software ticket planning by decomposing work into sub-tasks.

Community
Advanced
softsparksoftspark
161

pr

Create GitHub pull requests with pre-flight validation and automated commit summaries.

Community
Intermediate
KamiJeongKamiJeong

speckit-token-observability

Track token usage across Spec Kit orchestration and workflow phases.

Community
Advanced
aslaiiaslaii

gsd-fast

Execute trivial tasks directly without spawning subagents or creating PLAN.md files.

Community
Basic