naxxy/workspace-builder-skill
Developer tools이 페이지는 AI로 자동 번역되었습니다. 기술적인 세부 사항은 원문을 확인하세요. 원문 보기
An AI workspace builder skill based on the "Interpretable Context Methodology" by Jake Van Clief and David McDermott
개요
A Claude Code skill for creating, reviewing, updating, and auditing ICM-compliant workspaces. workspace-builder is a skill file for Claude Code. It implements the — a way of structuring a project folder so that a single Claude instance can orchestrate a multi-stage content or knowledge pipeline without a framework, without code, and without server infrastructure. The core idea: A properly layered workspace gives Claude exactly the context it needs at each stage, no more. Each stage has a contract (Inputs / Process / Outputs). The stage chain connects them. Claude reads only what it needs and produces exactly what the next stage expects. This skill builds and maintains those workspaces.
README
원문 README의 일부입니다. 원문 보기.
workspace-builder
A Claude Code skill for creating, reviewing, updating, and auditing ICM-compliant workspaces.
Skill build: 2026-05-21
What this is
workspace-builder is a skill file for Claude Code. It implements the Interpretable Context Methodology (ICM) — a way of structuring a project folder so that a single Claude instance can orchestrate a multi-stage content or knowledge pipeline without a framework, without code, and without server infrastructure.
The core idea: folder structure is agent architecture. A properly layered workspace gives Claude exactly the context it needs at each stage, no more. Each stage has a contract (Inputs / Process / Outputs). The stage chain connects them. Claude reads only what it needs and produces exactly what the next stage expects.
This skill builds and maintains those workspaces.
Why it exists
Working with Claude across a multi-stage pipeline without a structured workspace means:
- Claude re-reads everything at every session start
- Context bleeds between stages (research assumptions appearing in final output)
- There is no record of why the workspace was structured a particular way
- Recurring output problems are corrected by hand each run rather than fixed at the source
A well-structured ICM workspace addresses all of these. But building one correctly — with the right layer separation, stage contracts, routing table, reference material, and session persistence — requires knowing the methodology well. This skill encodes that knowledge and applies it consistently.
The 5-layer ICM hierarchy
Every workspace this skill produces follows the same layered structure:
| Layer | File | Token budget | Question it answers |
|---|---|---|---|
| 0 | CLAUDE.md (root) |
~800 tok | “Where am I?” |
| 1 | CONTEXT.md (root) |
~300 tok | “Where do I go?” |
| 2 | stages/NN_name/CONTEXT.md |
200–500 tok | “What do I do?” |
| 3 | stages/NN_name/references/ + _config/ |
500–2k tok | “What rules apply?” |
| 4 | stages/NN_name/output/ |
varies | “What am I working with?” |
Layer 3 is the factory. Layer 4 is the product. Layer 3 holds stable reference material (voice guides, domain knowledge, structural conventions) unchanged across every run. Layer 4 holds per-run working artifacts. Mixing them is the most common structural failure mode — the skill detects and fixes it.
Modes
The skill has five modes, selected automatically based on what you say:
| Mode | When to use | What it does |
|---|---|---|
| Setup | No workspace exists | Interviews you, generates a complete workspace |
| Review | “Is this well structured?” | Layer-by-layer assessment with prioritised fixes |
| Update | “Change X” or “Add a stage” | Minimum-viable-change to an existing workspace |
| Check | “Audit this” | Full anti-pattern and integrity check |
추천 도구
다른 키워드를 입력하거나 필터를 제거해 보세요.
설치
npx skillfish add naxxy/workspace-builder-skill