Summary

• Adds scoring/task_output_scorer.py — LLM-as-judge that scores pipeline task outputs against a 5-dimension rubric (Specificity, Actionability, Completeness, Internal Consistency, Conciseness)
• Adds scoring/score_run_task.py — CLI helper to score tasks from completed run directories
• Foundation building block for autonomous prompt optimization (fix: define column_property checks after class body to avoid NotImple… #94) and A/B testing promotion (fix: correct hallucinated MCP tool names in OpenClaw skill and proposals #59)

Test plan

• ast.parse() all new files
• Verify imports succeed in project venv
• derive_task_name() correctly parses task filenames
• DEFAULT_WEIGHTS sum to 1.0
• Manual test with a real LLM and run directory

🤖 Generated with Claude Code