Evaluate software engineering tasks with a 15-task benchmark and scoring rubric.
npx skills add https://github.com/baekenough/second-brain --skill harness-eval
Please help me install this Agent Skill. Skill: harness-eval Source: https://github.com/baekenough/second-brain/tree/main/.claude/skills/harness-eval Command: npx skills add https://github.com/baekenough/second-brain --skill harness-eval
Structured SE task evaluation uses a 15-task benchmark to provide objective scores for agent performance.
Run harness-eval with all benchmarks to generate the full results report.
High-intent search queries and answers about installing and using this skill.
A structured software engineering benchmark evaluates agent performance using a standardized 15-task harness suite. It applies a standardized rubric across API design, data modeling, authentication flow, logging, and configuration to generate objective per-task scores and an aggregate grade.
To run a software engineering benchmark evaluation, use the all preset to execute the full 15-task harness suite. This generates per-task scores, an aggregate grade, and exports the formatted results to the predefined .claude/outputs path.
Yes, you can use the quick preset for rapid software engineering benchmarking. It runs a faster subset of the 15-task suite to produce objective scores quickly, enabling rapid comparison of agent performance across standard tasks.
Yes, the benchmark evaluation integrates directly with the evaluator-optimizer workflow. It formats per-task scores and aggregate grades into outputs optimized for the evaluator-optimizer pipeline to consume and process.
The benchmark scoring rubric covers 15 tasks including API design, data modeling, authentication flow, logging, and configuration. It generates objective per-task scores and an aggregate grade for each.
Software engineering benchmark results are exported to the predefined .claude/outputs path. The evaluation outputs are formatted specifically for consumption by the evaluator-optimizer pipeline.
loveconnorModel security threats and privacy risks across the data lifecycle.
RB-chrismandichAggregate agent output logs into markdown performance dashboards.
Manavv007Implement iterative evaluation and refinement loops for AI agent outputs.
hiddink-aiEvaluate AI coding agents with 15 standardized software engineering tasks and quantitative scoring rubrics.
