eval-grader

Classify batch evaluation results and generate publication-ready markdown summary tables.

Updated Jul 2, 2026
One-click install
npx skills add https://github.com/SamyakJhaveri/loam --skill eval-grader
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: eval-grader
Source: https://github.com/SamyakJhaveri/loam/tree/main/seed/_research/skills/eval-grader
Command: npx skills add https://github.com/SamyakJhaveri/loam --skill eval-grader

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Automates the grading and classification of evaluation results from batch runs, producing defensible statistics for paper reporting.

Core Features & Use Cases

  • Load results from a results directory and apply exclusions from a known-issues list
  • Classify failure modes with a structured taxonomy and provide per-model statistics
  • Generate a summary table suitable for publication

Quick Start

Run the evaluator on a results directory to classify outcomes and generate a summary table

Frequently Asked Questions about eval-grader

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I grade evaluation results from batch runs for paper reporting?

Classifying model failure modes uses a structured taxonomy to categorize errors from evaluation runs. It diagnoses failure patterns by applying exclusions for known issues, ensuring defensible and accurate per-model statistics for reporting.

How do I exclude known issues from model evaluation statistics?

Excluding known issues from model evaluation statistics requires applying a known-issues list to loaded batch run results. This workflow filters out predefined failures before computing pass/fail statistics, ensuring reported outcomes reflect genuine model performance.

Can I generate a markdown summary table for experiment classification results?

Publication-ready summary tables require structured failure classification and computed pass/fail statistics from batch runs. By loading results, applying exclusions, and categorizing failures, the workflow produces defensible markdown tables suitable for papers.

Does this evaluation grading workflow support diagnosing failure patterns?

Diagnosing failure patterns is supported through structured failure mode classification. The workflow categorizes errors from batch runs using a defined taxonomy, helping identify and quantify specific failure types for model evaluation reporting.

What is the best way to quantify pass/fail statistics for model evaluation experiments?

Quantifying pass/fail statistics for model evaluation experiments is best handled by automating the grading workflow. It loads results, applies exclusions for known issues, classifies outcomes, and computes defensible statistics for paper reporting.