Documentation updates needed (from demo w Adrian)

[rhiannon-io]

1. Clarify the difference between LLM “tools” and Active Agent “actions”
   Adrian requested clear explanations when terminology overlaps or differs between Active Agent and LLM conventions (e.g. tools vs. actions).
   “If a tool in LLM world is different from a tool in Active Agent, I want you to tell us that.”

2. Add more complete working examples, including input/output samples
   Adrian emphasized:
   “If you give more examples on the documentation website with code snippets and what the result is, it’s a double plus. First for users, second for LLMs.”

3. Show rendered output and what views produce
   Justin notes this is often missing:
   “I’ve actually gotten feedback that I need to have some kind of output of what it does and what the view looks like when it’s rendered.”

4. Clarify how layout and instructions views work and interact
   Adrian had to confirm how those templates wrap prompt content and whether they apply globally or per agent. Worth a dedicated explanation in docs.

5. Include structured output patterns and example integrations
   Adrian asks about best practices for capturing structured output (like invoice fields), and Justin mentions an open PR related to this. Could be documented or supported more directly.

6. Show examples using before/after generation callbacks and on_stream
   These are powerful but undocumented features that would benefit from usage examples.

7. LLM doesn’t always receive instructions without an explicit message
   Prompting fails unless a message is passed, even if the instructions are defined. Needs more graceful fallback or clearer error.