Create and manage configuration-based workflow evaluations against test datasets.
npx skills add https://github.com/n8n-io/n8n --skill config-evals
Please help me install this Agent Skill. Skill: config-evals Source: https://github.com/n8n-io/n8n/tree/main/packages/%40n8n/instance-ai/skills/config-evals Command: npx skills add https://github.com/n8n-io/n8n --skill config-evals
Setting up workflow evaluations in n8n normally requires manually wiring on-canvas evaluation nodes, which is error-prone and disconnected from the workflow logic you actually want to score. This Skill attaches evaluations through the off-canvas evaluation-config API, pairing a workflow with a dataset and judge metrics without touching the canvas.
eval-config tool.correctness and helpfulness presets with judge credentials, models, and expression-based references to dataset columns and workflow outputs.Set up a correctness evaluation on my support agent workflow using the existing support eval dataset.
High-intent search queries and answers about installing and using this skill.
Use the `eval-config` tool with `action="create"`, supplying the workflow id, a start node after the trigger, an end node, a Data Table id, and one or more judged metrics. The config lives off-canvas and does not modify the workflow itself.
Correctness compares the produced answer to a ground-truth value from the dataset and requires an `expectedAnswer` expression. Helpfulness judges the answer against the user's query and requires a `userQuery` expression, with no ground truth needed.
An eval run replaces the workflow's trigger with a dataset-driven one, so the start node must already receive input from a trigger. Naming the trigger itself as the start node causes the eval run to fail to compile.
Expression fields like `actualAnswer`, `expectedAnswer`, and `userQuery` must begin with `=` so n8n evaluates the `{{ ... }}` template. Without the leading `=`, the string is stored verbatim and the judge scores the raw template text.
No. This Skill only handles configuration-based evaluations through the evaluation-config API. If the user asks for EvaluationTrigger or Evaluation nodes, build a config eval instead and explain that is how evaluations are set up.
jarodmengMap book units to answer pages and generate ground-truth JSON for pdf_file_manager.
jadecliCompute completeness, accuracy, and freshness scores to gate crawl output quality.
ricardorocheAutomate JSON-structured logging and OpenTelemetry tracing for AI/LLM pipelines.
7loroGenerate and edit JSON Canvas files with nodes, edges, and groups.