Thinking through what kinds of knowledge are worth storing in an entity knowledge base. Three types seem natural:
Facts — what exists. Properties of the environment, tools, or user. Semantic memory.
e.g. "GitHub traffic API requires admin access."
Guidelines — how to act. Rules that steer the agent toward better execution.
e.g. "Read a file before editing it."
Observations — why something is the way it is. Design rationale and decision context. Useful for explainability — answering "why did we do it this way?" rather than directing future action.
e.g. "Recall and learn logs are kept separate so they can be read independently to diagnose knowledge saturation."
The short version: facts are about what, guidelines are about how, observations are about why.
Interested in whether this taxonomy resonates and what others think about how these should be captured and surfaced differently.
Thinking through what kinds of knowledge are worth storing in an entity knowledge base. Three types seem natural:
Facts — what exists. Properties of the environment, tools, or user. Semantic memory.
e.g. "GitHub traffic API requires admin access."
Guidelines — how to act. Rules that steer the agent toward better execution.
e.g. "Read a file before editing it."
Observations — why something is the way it is. Design rationale and decision context. Useful for explainability — answering "why did we do it this way?" rather than directing future action.
e.g. "Recall and learn logs are kept separate so they can be read independently to diagnose knowledge saturation."
The short version: facts are about what, guidelines are about how, observations are about why.
Interested in whether this taxonomy resonates and what others think about how these should be captured and surfaced differently.