Add ROSE Feedback pattern

[devill]

Summary

• Introduces ROSE Feedback pattern for structuring feedback to AI agents
• ROSE = Risk, Observation, Solution, Expect
• Provides context so AI understands the principle, not just the instruction
• Solves Silent Misalignment anti-pattern

[devill]

@emilybache: Here is the pattern we discussed today. I'm curious of what you think of this write up.

@lexler: This pattern made me think if we need a separate recipes section for patterns that are more like Green Path style recipes then deeply consequential basic patterns. Essentially the patterns could be the key stone level behaviours, while the recipes are step by step instructions.

[lexler]

Feels like there's a generalization lurking here, around perferring generalizations to recipes that are too narrow

[devill]

@lexler I think the key is that specific recipes are more useful from a coaching perspective. Understanding and applying high level generalisations is a mapper level behaviour, while most of the time we need to get people from inspired to solo, where specific recipes are way more useful as a tool. That is why I'm suggesting to create a split there. There is the over all umbrella generalisation, but for each we need the specific recipes as kind of examples to teach to those who are not ready for the high level stuff just yet.