Best Evaluation agent skills
Measuring whether models, agents and prompts actually work — scores, traces and test harnesses. The 10 highest-ranked of the 10 agent skills tagged Evaluation in the Onei AI catalogue.
Everything tagged Evaluation: apps, skills and MCP servers Best Evaluation AI tools (apps), ranked
-
by Tencent
Tencent's skill-evolution harness: it rewrites a whole skill folder — SKILL.md, scripts and references together — and lands every decision as a real Git issue, PR and wiki entry you can review.
Editor's pick
-
by NVIDIA
Diagnoses wrong gradients in differentiable NVIDIA Warp programs by measuring first — comparing autodiff against finite differences on a shrunk reproduction before proposing any fix.
Also: Python · Editor's pick · Copied 1 time
-
by Anthropic
Create, edit, and optimize Agent Skills — scaffold new skills, refine existing ones, run evals, and tune descriptions for reliable triggering.
Copied 1 time
-
by Agnost AI
Walks a coding agent through wiring Agnost AI conversation analytics into a Python or TypeScript app — inspecting existing OpenTelemetry first and only adding an SDK when the traces are not usable.
-
by AMD
AMD's official skill for benchmarking vLLM/SGLang inference and driving GPU kernel optimization with Magpie and TraceLens trace analysis.
-
by Addy Osmani
Sends every non-trivial decision to a fresh-context adversarial reviewer before it stands — for production, security-sensitive or irreversible work.
-
by NVIDIA
NVIDIA's official skill for building synthetic datasets with NeMo Data Designer — describe the data you want and the agent writes a declarative generation pipeline, column by column.
Also: Python
-
by Anthropic
Anthropic's guide for building high-quality MCP servers end to end — research, tool design, implementation in TypeScript or Python, testing, and a 10-question evaluation suite.
-
by SerpApi
A protocol for measuring whether agents can discover and use your tool without being told about it — WITH/WITHOUT trials, binary scoring, and lift as the only metric. The subject under test is your docs, not the model.
Also: MCP
-
by Anthropic
Anthropic's official skill that appends 2-3 targeted follow-up questions to a substantive answer, prompting the reader to check facts, probe assumptions and spot missing context.
How this list is ranked
Ranked by 20 points for an editor's pick, plus 4 points per copy of the install command or SKILL.md up to 10, plus a point per 5 page views up to 100; ties go to the most recently updated. Discontinued skills are never ranked.