Model Skill Files (World Diff)
Model Skill Files is a concept for a publicly curated "world diff" that AI models can consume to stay current beyond their training cutoff. Rather than relying solely on ever-larger models, this approach gives any model access to structured world-state updates: what new products exist, which APIs have changed, what has been deprecated.
The Gap
No one currently maintains a community-curated, model-consumable knowledge update covering new products, API changes, naming updates, deprecations, and version bumps since a model's training cutoff. This is an open-source project opportunity.
Existing Partial Solutions
- Agent Skills (open standard): Anthropic created the SKILL.md pattern, adopted by Microsoft, OpenAI, and Cursor. These are task-specific skills, not world-state updates.
- llm-stats.com: Tracks model releases and updates but not in a model-consumable format.
- CCI (Claude Collective Intelligence): Captures problem/solution pairs discovered during Claude Code sessions — a narrow version of the concept.
- CLAUDE.md / system prompts: An ad hoc version of the idea, where developers manually note stale information in their system prompts.
Vision
A canonical, versioned, machine-readable file that any model can be given as context, enabling it to reason accurately about the current state of the world without retraining. This connects to The Context Engineering Prize — context engineering is not just about prompting technique but also about the quality and currency of the knowledge provided.
Status
Idea stage. Identified as a potential open-source project.
Related Ideas
Source: Voice note, February 9 2026