Skip to content
TrustedSkillsbeta
Back to all skills

Validate Evaluator

Checked
hamelsmu·vlatest·ai-ml·MIT

Checks evaluator outputs against ground truth data to identify biases, errors, and performance inconsistencies.

Repository

Install on your platform

Showing Claude Code. One command installs this skill into any of the agents below.

  1. 1

    Run this in your project

    terminal
    $ npx skills add hamelsmu/evals-skills --skill validate-evaluator -a claude-code

  2. Uses the open-source skills CLI and needs Node.js. Add -g to install it for your user instead of this project.

What we know about this skill

Publisher

hamelsmu

Taken from the source listing. We haven't verified who controls this account.

Install target

Live repository

Not pinned. Installing fetches whatever the repository holds at the time you run the command.

No human has reviewed this code. The checks below are a static scan of the files at one commit — nobody has run this skill or judged whether it works. A skill runs with whatever access you give your agent, so read the source before you install it.

Automated safety checks

scanned

This skill passed all 6 checks. We read its files at one commit and looked for the things that make a skill dangerous to run — not for whether it works.

  • Declares what it does: passed

    SKILL.md parses and declares a name and a description.

    SKILL.md exists, its frontmatter parses, and it declares a name and a description.

  • No undeclared network calls: passed

    Makes no network calls.

    Every host the skill dials — from its scripts, its config endpoints or the commands it tells the agent to run — is either well-known package and API infrastructure or declared in its own frontmatter. Links in documentation are not counted.

  • No obfuscated payloads: passed

    No encoded or obfuscated payloads.

    No base64, hex or compressed blob is decoded and then executed, and no encoded shell commands.

  • No credential access: passed

    Reads no keys, tokens, keychains or credential files.

    Neither the code nor the SKILL.md instructions read SSH keys, cloud credentials, keychains, token stores, browser cookie databases or dotfiles that hold secrets.

  • No pipe-to-shell installers: passed

    No pipe-to-shell or download-and-run installers.

    No `curl … | sh` pattern, in the code or in the instructions, that runs code fetched at run time.

  • Pinned to a commit: passed

    Scanned hamelsmu/evals-skills at 22418da2bf.

    The exact upstream commit SHA that was scanned is recorded, so the result refers to specific bytes.

Commit
22418da2bf
Manifest
skills/validate-evaluator/SKILL.md
Scanned
1 file, 202 lines
This is a static scan, not a review. It reads the skill's files; it does not run them, and it cannot tell you whether the skill is any good or whether the code does what its description says. A skill runs with whatever access you give your agent.
What each check looks for →

About this skill

What it does

This skill, Validate Evaluator, helps calibrate an LLM-based evaluator (judge) against human judgment. It guides you through splitting labeled data into training, development, and test sets, then iteratively refining the judge's prompt until its True Positive Rate (TPR) and True Negative Rate (TNR) exceed 90% on the development set. Finally, it allows for a final accuracy measurement using a held-out test set and provides a bias correction formula to apply to production data.

When to use it

  • You have built an LLM judge prompt and want to ensure its accuracy and reliability.
  • You need to compare the performance of your LLM judge against human evaluations.
  • You suspect biases or inconsistencies in your LLM evaluator's outputs.
  • You are preparing a production-ready LLM evaluation system.

Key capabilities

  • Data splitting into training, development, and test sets.
  • Measurement of True Positive Rate (TPR) and True Negative Rate (TNR).
  • Iterative refinement of the judge prompt based on development set performance.
  • Final accuracy assessment using a held-out test set.
  • Provides a bias correction formula for production data.

Example prompts

  • "Validate Evaluator: I have human-labeled data, split it into training, dev, and test sets."
  • "Validate Evaluator: Run the judge on my development set and tell me the TPR and TNR."
  • "Validate Evaluator: How do I adjust my LLM prompt to improve performance on the development set?"

Tips & gotchas

  • Requires a pre-existing LLM judge prompt (created using a skill like "write-judge-prompt").
  • Needs approximately 100 human-labeled traces per failure mode, with roughly equal numbers of Pass and Fail examples.
  • Labels must be provided by domain experts to ensure accuracy.