Prompt optimization: remove redundancy and shorten examples

[apenab]

Problem

Prompts are highly detailed and contain overlapping guidance (e.g., STRATEGY vs CRITICAL RULES), increasing token overhead.

Proposal

Optimize prompts by:

• Consolidating redundant instructions.
• Shortening or reducing examples where possible.
• Preserving all essential constraints and behavior-critical rules.

Expected Impact

• Estimated token reduction: ~10-15%
• Example target: 2.2M -> 1.9M tokens (~-13%)

Risk

• Low if optimization is careful and benchmarked.

Acceptance Criteria

• Prompt token count reduced by at least 10%.
• No significant quality regression on benchmark metrics.
• Diff/mapping of removed vs retained guidance is documented.
• A/B benchmark results are attached to the issue.