User Story

As Maya, in order to see quantitative benchmarks for the shaping model variants, I want a CLI runner that evaluates all three shaping models against the scripted intent suite.

Context

Story #59 shipped the shaping-commands evaluation kind and 6 scripted intents with expected outputs. The scoring logic is tested. What’s missing is a CLI command to actually call each model with the intents and record results.

Acceptance Criteria

• CLI command: bun run cli evaluate shaping run <variant-id>
• Runs all 6 scripted intents against the selected shaping variant
• Writes results to catalog/experiments/shaping-model-comparison/{variant}.json
• Generates markdown catalog page with metrics
• All three variants evaluated: bedrock-haiku, bedrock-sonnet, bedrock-opus
• Catalog pages updated with real metrics

Definition of Done

• Tests pass
• All three variants evaluated
• Catalog pages populated