Code Evaluation Harness

What it does

This skill provides an evaluation harness that spawns a separate evaluator agent to independently assess code artifacts across four axes: functional accuracy, code quality, originality, and usability & security. The evaluator produces a numeric score and a written report (EVAL_REPORT.md), with recommendations and a pass/conditional/fail verdict.

When to use it

Use when you need an automated, repeatable assessment of generated or submitted code — for grading, CI checks, or quality gates. Trigger on keywords like eval, quality score, code evaluation, or when a project requires an impartial scoring pass before merge or release.

What's included

• Scripts: none embedded (has_scripts=false)
• References: no references directory (has_references=false)
• Instructions: procedural steps to spawn an evaluator subagent with a fixed prompt, collect the EVAL_REPORT.md, and optionally run pass@k consistency checks by repeating evaluations to measure score variance.

Compatible agents

Designed for multi-agent or subagent-capable systems (agents that can spawn evaluator subagents), and helpful for developer-assistants and CI-integrated bots (Claude-style agents, acp harnesses, other orchestrators).