Verified
Automated LLM benchmark evaluation
Best for: Ask your agent: "Use /evaluation-harness to execute and coordinate this task."
npx -y skills add jgamaraalv/delivery-loop --skill evaluation-harness --agent claude-codePrevent hallucinations, jailbreaks, and prompt regressions. These specialized subagents build automated testing harnesses, safety guardrails, and deterministic evaluation rubrics.
4 Verified
Claude, Cursor, Windsurf
1-Click Install
Weekly Benchmarks
Ranked by real-world developer adoption, benchmarked reliability, security posture, and documentation depth.
Automated LLM benchmark evaluation
npx -y skills add jgamaraalv/delivery-loop --skill evaluation-harness --agent claude-codeContent moderation & safety filters
npx -y skills add jgamaraalv/delivery-loop --skill guardrails-safety-filter-builder --agent claude-codeHallucination & prompt error diagnosis
npx -y skills add jgamaraalv/delivery-loop --skill llm-debugger --agent claude-codePrompt regression & diff testing
npx -y skills add jgamaraalv/delivery-loop --skill prompt-regression-tester --agent claude-code| Rank | Name | Stars / Popularity | Best Use Case | Details |
|---|---|---|---|---|
| #1 | LLM Evaluation Harness | Curated Pick | Ask your agent: "Use /evaluation-harness to execute and coordinate this task." | View → |
| #2 | Guardrails & Safety Builder | Curated Pick | Ask your agent: "Use /guardrails-safety-filter-builder to execute and coordinate this task." | View → |
| #3 | LLM Failure Debugger | Curated Pick | Ask your agent: "Use /llm-debugger to execute and coordinate this task." | View → |
| #4 | Prompt Regression Tester | Curated Pick | Ask your agent: "Use /prompt-regression-tester to execute and coordinate this task." | View → |
Our evaluation methodology ensures every listed tool satisfies production engineering standards.
Robust defense against prompt leakage, jailbreaking, and unauthorized tool calls.
Objective, repeatable scoring rubrics using LLM-as-a-judge and ground truth datasets.
Updating a system prompt to fix one edge case often breaks existing behavior. The regression tester runs automated test suites against previous outputs to catch unintended regressions.