AI

ai-eval-review

Evaluates AI performance evidence to confirm accuracy, grounding, and valid tool usage.

Install

mkdir -p .claude/skills/ai-eval-review && curl -L -o skill.zip "https://agentskills.codes/api/skills/download/12879" && unzip -o skill.zip -d .claude/skills/ai-eval-review && rm skill.zip

Installs to .claude/skills/ai-eval-review

Activation

This is the description your AI agent reads to decide when to run this skill — the better it matches your request, the more reliably it fires.

Review AI evaluation evidence for grounding, hallucination risk, tool assumptions, and whether results actually support the claimed behavior.
141 charsno explicit “when” trigger
Intermediate

Key capabilities

  • Compare evaluation results to stated behavior
  • Check for hallucination risk in AI output
  • Verify retrieval and tool assumptions
  • Confirm eval exercises claimed behavior
  • Identify gaps missed by automated scores

How it works

This skill reviews AI evaluation evidence by comparing results to stated behavior, checking for risks, and confirming the evaluation's scope.

Inputs & outputs

You give it
AI evaluation evidence, design docs, tests, and prompt/scenario
You get back
Pass/fail/partial status, notes on hallucination/grounding, and follow-up recommendations

When to use ai-eval-review

  • Reviewing AI evaluation metrics
  • Checking for model hallucinations
  • Validating test evidence
  • Grounding AI output claims

About this skill

AI Eval Review

Purpose

Decide whether evaluation evidence is strong enough to support the claimed AI behavior.

Read

  • .codex/project-context.md
  • relevant design docs
  • relevant tests and validation output
  • the prompt or eval scenario under review

Do

  1. Compare eval results to the stated behavior and grounding.
  2. Check hallucination risk, retrieval/tool assumptions, and prompt boundaries.
  3. Confirm the eval exercises the behavior it claims to prove.
  4. Identify gaps that automated scores may miss.
  5. Recommend prompt, design, test, or validation updates as separate findings.

Outputs

  • Pass/fail/partial status per evaluation scenario.
  • Notes on hallucination checks, grounding, and tool usage.
  • Follow-up recommendations.

Rules

  • Do not approve plausible model output without grounding evidence.
  • Do not conflate eval coverage with complete behavior validation.
  • Do not change requirements while reviewing eval evidence.

When not to use it

  • When plausible model output is acceptable without grounding evidence
  • When eval coverage is considered complete behavior validation
  • When requirements should be changed during review

Limitations

  • Does not approve plausible model output without grounding evidence
  • Does not conflate eval coverage with complete behavior validation
  • Does not change requirements while reviewing eval evidence

How it compares

This workflow provides a structured review of AI evaluation evidence, focusing on grounding and hallucination risks, which goes beyond automated scoring alone.

Compared to similar skills

ai-eval-review side by side with the closest alternatives in the catalog.

SkillInstallsUpdatedSafetyDifficulty
ai-eval-review (this skill)02moNo flagsIntermediate
mlops-validation26moNo flagsIntermediate
python-testing-patterns772moReviewIntermediate
dependency-upgrade265moReviewIntermediate

Try saying

Example prompts that trigger this skill in your AI assistant.

You might also like

Search skills

Search the agent skills registry