eval-audit-and-sweep

eval-audit-and-sweep is a skill for Claude Code from anthropics/cwc-workshops. It costs 0 tokens per session (692 once invoked), scanned A, original, Apache-2.0.

A guide for checking the quality of an existing test suite for language models and comparing models or settings through repeated evaluations. An evaluation is a set of tests used to measure how reliably a model performs a task.

In plain words
What is it for?
Auditing LLM evaluations, testing different models and inference settings, running parameter comparisons, and making evidence-based model choices.
Why use it?
It helps identify unreliable tests and find a suitable balance between answer quality, cost, and speed.

Skill for Claude Code ✓ vendor

Written for Claude Code: installed under .claude/. Also seen: names the AskUserQuestion tool.

Good fit Auditing LLM evaluations, testing different models and inference settings, running parameter comparisons, and making evidence-based model choices.

Compare 6 skills from other repositories ↓
Install with agentmods
npx agentmods add skills/anthropics/cwc-workshops/eval-audit-and-sweep
About the project

CWC Workshops is a collection of materials from Anthropic-run workshops on building and evaluating AI-assisted coding workflows. The workshops cover model selection, multi-agent systems, managed agents, and product development with coding agents. The catalogue entries are examples and teaching materials from those workflows.

anthropics/cwc-workshops · 2,051 stars · on GitHub

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

Any agent
npx skills add anthropics/cwc-workshops --skill eval-audit-and-sweep
Clone the repo
git clone --depth 1 https://github.com/anthropics/cwc-workshops

Made for: Claude Code.

Wrote this? Show the measurements

A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.

agentmods badge for eval-audit-and-sweep

README.md
[![agentmods](https://agentmods.dev/badge/skills/anthropics/cwc-workshops/eval-audit-and-sweep/github.svg)](https://agentmods.dev/skills/anthropics/cwc-workshops/eval-audit-and-sweep)
Your own site
<a href="https://agentmods.dev/skills/anthropics/cwc-workshops/eval-audit-and-sweep"><img src="https://agentmods.dev/badge/skills/anthropics/cwc-workshops/eval-audit-and-sweep/github.svg" alt="Measured on agentmods" height="20"></a>

Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.

agentmods 80×15 button for eval-audit-and-sweep

Your own site · 80×15
<a href="https://agentmods.dev/skills/anthropics/cwc-workshops/eval-audit-and-sweep"><img src="https://agentmods.dev/badge/skills/anthropics/cwc-workshops/eval-audit-and-sweep.svg" alt="Reviewed on agentmods" width="80" height="20"></a>
Per session 0 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 692 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. A grade says what 26 rules found in the file — not that it is safe. Third-party audits
  • NVIDIA SkillSpector pass 7 Sept 2026
How audits are shown
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5.1 $0.00000 $0.00692
Opus 5 $0.00000 $0.00346
Sonnet 5 $0.00000 $0.00138
Haiku 4.5 $0.00000 $0.00069

Measured 11d ago against content hash e8f8a7e3ae16, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-11, from the pricing page.

Security

Grade A, and why

eval-audit-and-sweep scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 11d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

rightmodel/.claude/skills/eval-audit-and-sweep/SKILL.md · 28 lines

How it starts

The opening of the file, as written. The whole thing — 28 lines — stays where its author put it; the contents beside it link to each section on GitHub.


name: eval-audit-and-sweep description: This skill should be used when a user wants to (a) audit the quality and reliability of an existing LLM evaluation suite, or (b) determine which Claude model and inference parameters give the best quality-per-dollar and quality-per-second for their specific task by running a parameter sweep over that eval. Applicable to any eval framework (custom harnesses, tau-bench, inspect-ai, pytest-based, etc.) since the guidance is framework-agnostic.

Eval Audit and Sweep

This skill is an example exercise for the "Picking the Right Model" workshop during Code with Claude. It is a two-phase playbook for getting trustworthy cost-quality numbers out of an existing LLM eval. Phase 1 audits the eval for common reliability issues. Phase 2 wraps it in a model/parameter sweep and produces a recommendation. The phases are independent: a user may ask for only the audit, only the sweep, or both.

The skill contains guidance and example snippets only; it ships no runnable scripts, because every eval framework is structured differently. Claude is expected to read the user's eval code, apply the principles in the reference files, and write whatever glue code that specific codebase needs.

How to use this skill

  1. Locate the eval. Find the golden set, the judge/scoring function, and the entrypoint that runs one full pass. If the user has not pointed at a specific directory, ask.

  2. Decide which phase(s) apply. If the user says "is my eval any good" or "review my eval," run Phase 1. If they say "which model should I use" or "what's the cheapest config that still passes," run Phase 2. If they say both or it is ambiguous, run Phase 1 first (a sweep over a broken eval produces misleading numbers).

  3. Read the relevant reference file before acting:

    • references/audit.md for Phase 1: the health-check checklist (task design, harness design, metrics hygiene, and grader design including LLM-judge biases) and how to report findings in measured, non-dogmatic language.
    • references/sweep.md for Phase 2: choosing the grid, instrumenting per-cell metrics, plotting, and stating a one-sentence recommendation. That file links out to harness-specific companions where they exist.

Read the full file on GitHub · 28 lines

Files

What ships with it

3 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 11d ago First seen · 28 lines · 0 tokens per session scan A e8f8a7e3ae16

Subscribe to this mod's changes

eval-audit-and-sweep is a skill published in the GitHub repository anthropics/cwc-workshops (2,051 stars, last pushed 14d ago), licensed Apache-2.0. It costs nothing until one of its globs matches a file; then it loads 692 tokens. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.

Related

Other skills, from other repositories

research-engineer

An uncompromising Academic Research Engineer. Operates with absolute scientific rigor, objective criticism, and zero flair. Focuses on theoretical correctness, formal verification, and optimal implementation across any required technology.

davila7/claude-code-templates · 43 tokens

tika-eval-compare

Compare extracts from two Tika builds over a corpus to detect regressions in content, encoding, exceptions, and embedded-document handling. Use for "compare before/after extracts", "eval this change against the corpus".

apache/tika · 50 tokens

neuron-evaluation-engineer

Create and run AI evaluations with datasets, assertions, and output drivers in Neuron AI. Use this skill whenever the user mentions evaluation, testing AI systems, creating evaluators, dataset-driven testing, assertion-based validation, or wants to measure AI system performance. Also trigger for tasks involving…

neuron-core/neuron-ai · 77 tokens

jetson-validate-image

Use after jetson-flash-image to run static BSP checks, on-target smoke/regression tests on a flashed DUT, or both. Not for build or flash steps. Triggers: validate bsp, on-target validation.

NVIDIA/skills · 50 tokens

atmos-validation

Validate Atmos projects, components, arbitrary JSON Schema inputs, EditorConfig, and GitHub Actions; use affected-file selection and native CI annotations.

cloudposse/atmos · 31 tokens

skill-benchmark

Benchmark AI skill effectiveness by measuring implementation quality against legacy constraints.

HoangNguyen0403/agent-skills-standard · 16 tokens