SkillStack

The description states that SkillStack is a research prototype exploring continual learning in language models through the use of small frozen LoRA "skill frames" selected via a router. During OpenAI Build Week 2026, the author extended this with a reproducible routing audit to validate chain-style routing behavior. The project is self-reported as a hackathon submission and not independently verified.

The description states that SkillStack is a continual-learning research prototype built during OpenAI Build Week 2026. It explores how language models can learn new skills without retraining large monolithic models by using small frozen LoRA "skill frames" and selecting/composing them through a router.

The author notes that the original project was already a continual-learning research prototype before the hackathon, and during Build Week they added a reproducible routing audit to validate chain-style routing behavior. The new work includes:

• A CPU-only routing audit script
• True parent_id traversal validation
• Checks for missing parents and cycles
• Deterministic comparison between routing modes
• Unit tests for audit logic
• JSON and Markdown audit outputs
• Semantic-shift benchmark schema validator

The author clarifies that the demo output is synthetic, not a production-level benchmark claim.

The description states that SkillStack was inspired by a research question: whether language models can keep learning new skills without constantly retraining one large monolithic model. The project explores an alternative direction using small frozen LoRA "skill frames" and router selection/composition.

During OpenAI Build Week 2026, the author extended the original research prototype with a routing audit to improve transparency and testability of the routing logic. The author notes that GPT-5.6 was used as a research reviewer to identify weaknesses in the existing project, specifically around routing hierarchy validation.

The positioning appears to be that SkillStack is a framework for experimenting with continual learning, adapter composition, routing, compression, and AI-assisted ML research workflows. However, there is no evidence of any commercial positioning or claims about market adoption or product readiness beyond the author's own description.

Not evidenced. The description does not state what specific customer segments or ideal customer profiles (ICPs) this project targets. There are no mentions of end-users, adopters, or target industries.

Not evidenced. The description makes no claims about business models, pricing structures, revenue streams, or monetization strategies beyond the author's own account of a research prototype.

The description states that SkillStack was built using:

• Codex
• GPT-5.6 (used for research review)
• GitHub
• Python
• PyTorch
• Machine learning
• Natural language processing
• LoRA
• JSON
• Unit testing

The author notes that the Build Week extension was implemented using Codex as an implementation partner and included:

• Routing audit script
• Semantic-shift validator
• Unit tests
• Documentation
• Reproducible output files

The project is described as having a CPU-only pipeline for reproducibility, avoiding GPU dependencies, long runtimes, and fragile dependencies. The author emphasizes that the goal was not to fake a large breakthrough but to create a clean, honest, and testable extension.

Not evidenced. There is no evidence of revenue, customers, user adoption, or product-market fit beyond the author's own description. The project is described as a research prototype with no mention of any traction indicators such as users, downloads, engagement metrics, or market validation.

Not evidenced. The description does not provide information about competitors, market positioning, or competitive landscape. No mentions of existing solutions, market size, or competitive advantages are included.

• No commercial traction: The project is described as a research prototype with no evidence of revenue, customers, or product-market fit.
• Self-reported only: All information comes from the author's own account without independent verification.
• Hackathon context: The work was done for a hackathon submission, suggesting it may be experimental rather than production-ready.
• Limited scope: The extension focused on improving routing audit logic rather than implementing core functionality.
• Synthetic outputs: The demo output is described as synthetic, not a real benchmark claim.
• Single-person team: The project was built by one person (Piotr Gawron), which may limit scalability and development capacity.

1. What specific research problems or use cases does SkillStack aim to solve?
2. How does this prototype differ from existing continual learning approaches in the market?
3. What are the technical limitations of the current implementation that would need to be addressed for production use?
4. Are there any plans to commercialize this technology, and if so, what is the business model?
5. What validation or testing has been done beyond the Build Week extension?
6. How does the routing audit improve upon existing methods for validating chain-style routing behavior?
7. What are the key assumptions underlying the project's approach to continual learning?
8. How would you scale this research prototype into a production system?