xiry.

AI · GitHub repo

Interpretable Context Methodology (ICM)

Folder structure as agent architecture for multi-stage AI workflows

From the Xiry library · Updated 3 October 2026

Visit github.com ↗

Interpretable Context Methodology (formerly Model Workspace Protocol) replaces complex multi-agent orchestration frameworks with a structured filesystem of Markdown files. By organizing workflows into numbered stage directories, a single AI agent loads only the necessary context per stage, keeping prompt overhead low. Every intermediate output remains a plain text file, allowing users to easily inspect, audit, and edit outputs between steps.

What it's for

  • Convert raw documents into presentation slide decks via automated pipelines
  • Turn content ideas and research notes into scripts and Remotion animation code
  • Generate voice-driven explainer videos by coordinating narration generation, beat transcription, and video rendering
  • Build custom multi-stage production systems for specialized business workflows without writing framework orchestration code

Features

  • Five-layer context hierarchy to prevent prompt bloat and keep token usage low
  • Numbered stage folders that establish clear execution order and separation of concerns
  • Markdown stage contracts defining explicit inputs, processing steps, and output files
  • Human-in-the-loop edit surfaces where outputs of one stage can be manually reviewed before the next runs
  • Integrated checkpoints and self-audit quality gates for creative decisions
  • Config-once factory setup for brand voice, styling rules, and reusable domain skills

Pricing

Open source. Free and open source (MIT license)

Details

Type
GitHub repo
Category
AI
Topic
Agent Orchestration
Open source
Yes
Platforms
Command line, other
For
Developers and AI practitioners building repeatable, human-supervised generative pipelines with Claude Code or similar agentic tools
Alternative to
CrewAI, LangChain, AutoGen
Tags
agents, claude, context engineering, markdown, orchestration, prompt engineering, workflow

Why it was saved

  • @lostandlucky, 23 February 2026: Connects and automates complex multi-step generative tasks (like generating course content, slides, and animations) into a repeatable pipeline.

Compare

Related

Updated