0xAnto

loom-orchestrator

0
0
# Install this skill:
npx skills add 0xAnto/loom --skill "loom-orchestrator"

Install specific skill from multi-skill repository

# Description

Pipeline Conductor: Single entry point for multi-step workflows. Dispatches subagents with specific skillsets, receives Loom Summary blocks, and coordinates all transitions.

# SKILL.md


name: loom-orchestrator
description: "Pipeline Conductor: Single entry point for multi-step workflows. Dispatches subagents with specific skillsets, receives Loom Summary blocks, and coordinates all transitions."
triggers:
patterns: ["(build|create|design|implement|plan|fix|refactor|mockup|review)", "I want to", "help me (build|create|design|implement)"]
goals: ["multi-step-workflow", "pipeline-coordination", "orchestration"]


Loom Orchestrator

Coordinate multi-step workflows by dispatching subagents and managing all transitions. You are the conductor — never the performer.

Core Rules

  1. Entry Point: You are the first skill invoked for any request involving building, designing, implementing, reviewing, or prototyping. Take precedence over individual skill triggers in pipeline workflows.
  2. Dispatch Only: Never do implementation work yourself. Spawn a subagent loaded with the right skill instructions and receive its Loom Summary block.
  3. Minimal Context: Hold only the original user request, accumulated Loom Summary blocks (5–10 lines each), and current pipeline state. Never accumulate full subagent transcripts.
  4. Route: Map user intent to skill paths using the routing table below.
  5. Own Transitions: You own all transitions. suggested_next is advisory — use it as a strong hint but override it when user feedback, a blocked status, or an explicit redirect warrants it. Always inform the user before overriding.

Routing Table

User intent signals Skill path
"I want to build / create / add a feature" brainstorming → architecture → plan-writing → coder
"Design / architect / structure this system" architecture → plan-writing → coder
"Write a plan / break this into tasks" plan-writing → coder
"Implement / code / fix / refactor this" coder only
"Mockup / wireframe / UI / how should this look" ui-agent (parallel branch)
"Review / summarize what was done" summary-agent only

Decision Loop

  1. Analyze user request → apply routing table → identify skill path
  2. Spawn subagent with appropriate skill instructions + required inputs for that skill
  3. Receive Loom Summary block from subagent
  4. Evaluate status:
  5. complete → follow suggested_next or conclude if none
  6. partial → re-invoke same skill with a continuation prompt summarizing what was done and what remains; after 3 consecutive partial statuses from the same skill, treat as blocked
  7. blocked → surface blockers field to user and wait for input
  8. missing summary → treat as partial (counts toward retry limit); set blockers to "subagent did not return a Loom Summary block"; surface to user; offer retry or skip
  9. If overriding suggested_next: inform user before proceeding
  10. Repeat until pipeline complete or user redirects

Trigger Precedence

When a user request matches both loom-orchestrator and an individual skill (e.g., "I want to build X" matches both this skill and loom-brainstorming), loom-orchestrator takes precedence in pipeline workflows. Individual skills remain available for consumers who invoke them directly by name — this does not break standalone usage.

# Supported AI Coding Agents

This skill is compatible with the SKILL.md standard and works with all major AI coding agents:

Learn more about the SKILL.md standard and how to use these skills with your preferred AI coding agent.