Compare two variants in parallel with a model grader and structured report.
npx skills add https://github.com/baleen37/bstack --skill eval-harness-baleen37
Please help me install this Agent Skill. Skill: eval-harness Source: https://github.com/baleen37/bstack/tree/main/plugins/me/skills/eval-harness Command: npx skills add https://github.com/baleen37/bstack --skill eval-harness-baleen37
Streamlines and standardizes variant comparisons by running two options in parallel and using a model-based grader to produce an objective, comparable assessment.
Provide a task and two variants with inputs to initiate automated parallel evaluation and grading.
High-intent search queries and answers about installing and using this skill.
You can compare code, LLM prompts, CLI commands, or custom workflows by providing the TASK, VARIANT_A, VARIANT_B, INPUTS, and Evals fields to initiate automated parallel evaluation and grading.
Parallel variant evaluation uses two subagents to execute different options simultaneously with identical inputs, followed by a single model-based grading pass that produces an objective, comparable assessment with per-criterion verdicts.
Yes, the harness supports variant types including code, LLM, command, and custom workflows, using a unified input specification to ensure identical inputs across both variants for fair, apples-to-apples comparison.
The best way is using a structured harness that anonymizes variants for fair judging, runs them in parallel with identical inputs, and applies a model grader to evaluate observable outputs across defined criteria.
Yes, you must provide the TASK, VARIANT_A, VARIANT_B, INPUTS, and Evals fields to initiate the automated parallel evaluation and grading process for accurate variant comparison.
alimCreate, improve, and measure AI skills through sub-agent workflows.
prestariusAudit Claude Code skills for overlap, freshness, and trigger integrity.
modu-aiManage Git worktrees for isolated parallel SPEC development in MoAI-ADK workflows.
OmidZamaniEvaluate DSPy programs with custom metrics and parallel execution.
