- Author
- Brian Carpio, OutcomeOps
- Published
- Undated
- Kind
- GitHub README for a reference implementation, from a company
- Original
- github.com/outcomeops/context-engineering
- Subject
- Harnesses and agent loops
What it is
Brian Carpio at OutcomeOps wrote the repo as a companion to a glossary page. Each numbered folder implements one component against a Spring PetClinic example. The README also lists case studies and further reading.
What it says
- The repo defines context engineering as designing, retrieving and injecting the information an AI needs.
- It names five components: corpus, retrieval, injection, output and enforcement.
- A system with only the first three components is a RAG system.
- The output and enforcement layers make results reviewable and governable.
- The examples run on Amazon Bedrock with Claude and Titan embeddings.
- One example claims three ADRs changed generated code on a Spring PetClinic codebase.
- The README says the harder shift is organizational.
What evidence it gives
The README points to example code and one A/B comparison on two public branches. It lists case studies with figures, such as production in 6 days. The README gives no method or data for those figures.
Interests and limits
OutcomeOps sells related products, and the README links its commercial site and case studies. Only the README was read, not the component folders or linked case studies. The example code was not run. The claims about results are the company's own.