Stocks-Trend-Screener Evaluator

Evaluate one run of the stocks-trend-screener skill by running an actor then scoring with a blind judge. Append results to eval.csv.

Role

You are an evaluation orchestrator. Your job is to run the screener, get a result, score it against the rubric, and record the score. You do NOT judge the output yourself — you delegate scoring to an independent subagent.

Rubric file

.agents/skills/stocks-trend-screener/evals/RUBRIC.md — 6 dimensions, 0–5 each:

1. source_grounding — did it read real journalism, or hallucinate/speculate?
2. non_obvious_discovery — did it map past the obvious leader to a hidden beneficiary?
3. skeptic_discipline — did it kill candidates that fail (already priced, no catalyst, no named risk)?
4. actionability — are finalists concrete enough for multi-lens-quorum to judge?
5. quorum_routing — did it route to quorum, not self-decide buy/sell?
6. prescreen_usage — did it use the quantitative scanner as Step 1 to direct reading?

Eval cases

• Train cases: .agents/skills/stocks-trend-screener/evals/cases/train/
• Holdout case: .agents/skills/stocks-trend-screener/evals/cases/holdout/