orch-ui

Orchestrates multi-step UI work cycles by running each step in an isolated subagent.

Updated Apr 9, 2026
One-click install
npx skills add https://github.com/muselesscreator/ui-skills --skill orch-ui-muselesscreator
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: orch-ui
Source: https://github.com/muselesscreator/ui-skills/tree/main/orch-ui
Command: npx skills add https://github.com/muselesscreator/ui-skills --skill orch-ui-muselesscreator

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Running a full UI feature workflow (analyze, plan, implement, validate, cleanup, commit) in one context pollutes the conversation with every step's artifacts and makes failures hard to isolate. This Skill coordinates the whole cycle while keeping each step in its own isolated subagent, handing off results through files so the orchestrator stays thin. ## Core Features & Use Cases - Cycle-type selection with confirmation: Enumerates available cycle definitions (feature-cycle, eli-feature-cycle, simple-task), guesses the right one from your task, and confirms with you before running anything. - Isolated step execution: Each step runs in its own subagent with configurable model tier (haiku/sonnet/opus) and agent type, returning only a five-line status report (STATUS, ARTIFACT, SUMMARY, FOLLOWUP, DECISIONS). - Failure gates and bounded remediation: Steps with stop_on_fail halt the cycle on failure; steps can declare bounded auto-remediation (e.g. re-run impl against a failing validation report up to N rounds), and BLOCKED steps pause for human decision-memo resolution. - Use Case: Run /orch-ui feature-cycle add a dark-mode toggle to settings — the orchestrator confirms the cycle, then sequences analyze → plan → impl → validate → cleanup → commit → retro, each in a fresh subagent, and prints a final status table. ## Quick Start Ask the assistant to run /orch-ui with a short description of the UI feature or change you want taken through the full plan-to-commit cycle.

Frequently Asked Questions about orch-ui

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

FAQPage Schema
How do I run a full UI feature cycle from plan to commit?

Invoke /orch-ui with a task description, confirm the suggested cycle-type (e.g. feature-cycle), and it runs analyze, plan, impl, validate, cleanup, commit, and retro in order. Each step executes in its own isolated subagent and hands off through files in the skill-output directory.

What is the difference between feature-cycle and simple-task?

feature-cycle runs the full analyze, plan, validate, cleanup, commit, retro loop with opus-tier planning and validation gates. simple-task runs only impl, cleanup, and commit on sonnet and haiku, intended for small already-scoped changes like copy fixes or one-file edits.

What happens when a validation step fails during a cycle?

A failed step with stop_on_fail halts the cycle unless it declares bounded remediation. With remediation configured, the orchestrator spawns the remediation skill against the failing artifact and re-runs the gate, up to the configured round cap, before halting.

Can a cycle use skills defined inside my own repository?

Yes. Steps with scope set to repo resolve the skill from the triggering repo's .agent/skills or .claude/skills directory. If the repo does not provide that skill, the step is skipped automatically rather than failing.

Why does a cycle stop with BLOCKED status instead of failing?

BLOCKED means an isolated subagent raised decision memos instead of guessing, since subagents cannot ask questions directly. The orchestrator presents each open memo to you, records your answers, then re-runs the same step and continues the cycle.

When should I not use the simple-task cycle?

Avoid simple-task when the change spans multiple files or domains, the approach is undecided, or the result needs a behavioral validation gate beyond lint and type checks. Those cases need feature-cycle's analyze, plan, and validate steps.