Hi, and thanks for publishing this context engineering guide.
It has been very useful to have a structured treatment of the topic.
I wanted to propose one external resource that might complement your material on
context quality, failure modes, and debugging:
Short summary:
- An open source, MIT-licensed problem map of 16 recurring failures in real-world
RAG and context-heavy LLM systems (retriever issues, chunking, poisoning,
prompt injection, vector store skew, etc).
- Each problem page contains:
- Observable symptoms that engineers see in logs or user reports
- Likely root causes across retrieval, data prep, and prompt design
- Concrete “minimal fix” playbooks that do not require changing providers
- It is designed to be used with any provider and any stack.
The only requirement is that the system uses retrieval or long-context prompts.
How it might fit your repo:
- Your work explains how to think about context engineering.
WFGY 16 Problem Map gives readers a checklist they can use on a live system to
locate which part of the context pipeline is actually failing.
- It can be referenced as:
- A “Further reading” link in a section on RAG or context debugging, or
- An example of a practical, failure-mode based framework for context problems.
If this does not align with your scope or style, please feel free to ignore this.
In any case, thank you for making high quality context engineering material public.
Hi, and thanks for publishing this context engineering guide.
It has been very useful to have a structured treatment of the topic.
I wanted to propose one external resource that might complement your material on
context quality, failure modes, and debugging:
https://github.com/onestardao/WFGY/tree/main/ProblemMap/README.md
Short summary:
RAG and context-heavy LLM systems (retriever issues, chunking, poisoning,
prompt injection, vector store skew, etc).
The only requirement is that the system uses retrieval or long-context prompts.
How it might fit your repo:
WFGY 16 Problem Map gives readers a checklist they can use on a live system to
locate which part of the context pipeline is actually failing.
If this does not align with your scope or style, please feel free to ignore this.
In any case, thank you for making high quality context engineering material public.