evals

evals is a command for Claude Code from artemnovichkov/skills. It costs 27 tokens per session (603 once invoked), scanned A, original, MIT.

A command that builds a regression test suite for an AI agent's configuration, including its instructions, skills, and hooks, and runs that suite in continuous integration. Continuous integration is an automated service that checks changes when code is submitted.

In plain words
What is it for?
Use it to collect cases from commits, corrections, and incidents, define acceptance checks, and run them automatically in CI.
Why use it?
Agent instructions can stop working after edits without causing ordinary application tests to fail. These evaluations preserve real past tasks as checks for future changes.

Command for Claude Code

Written for Claude Code: allowed-tools in frontmatter. Also seen: mentions CLAUDE.md.

Runs only inside its plugin — its command needs a path that Claude Code sets for a plugin’s own hooks and for nothing else. Install the plugin, not this.

Part of the sdlc plugin — 1 skill, 8 commands, 1 agent, 1 hook shipped together

Good fit Use it to collect cases from commits, corrections, and incidents, define acceptance checks, and run them automatically in CI.

Compare 6 commands from other repositories ↓
Install

Getting it into your agent

This one installs as part of its plugin. Adding the marketplace and installing the plugin brings it with everything else the plugin ships.

Claude Code
/plugin marketplace add artemnovichkov/skills
Claude Code
/plugin install sdlc

Made for: Claude Code.

Or install sdlc, the plugin that ships this one along with the rest of its 1 skill, 8 commands, 1 agent, 1 hook.

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 evals

README.md
[![agentmods](https://agentmods.dev/badge/commands/artemnovichkov/skills/evals/github.svg)](https://agentmods.dev/commands/artemnovichkov/skills/evals)
Your own site
<a href="https://agentmods.dev/commands/artemnovichkov/skills/evals"><img src="https://agentmods.dev/badge/commands/artemnovichkov/skills/evals/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 evals

Your own site · 80×15
<a href="https://agentmods.dev/commands/artemnovichkov/skills/evals"><img src="https://agentmods.dev/badge/commands/artemnovichkov/skills/evals.svg" alt="Reviewed on agentmods" width="80" height="20"></a>
Per session 27 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 603 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.
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 $0.00027 $0.00603
Opus 5 $0.00014 $0.00302
Sonnet 5 $0.00005 $0.00121
Haiku 4.5 $0.00003 $0.00060

Measured yesterday against content hash 17f432b32080, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-08, from the pricing page.

Security

Grade A, and why

evals 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 yesterday.

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.

skills/sdlc/commands/evals.md · 56 lines

How it starts

The opening of the file, as written. The whole thing — 56 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Stage 4B — Continuous evals for agent configuration

CLAUDE.md, skills, and hooks are code that steers every session. They regress silently. Evals are the regression suite for them.

Workflow

1. Collect real cases

Do not invent prompts. Pull 20–50 tasks the repo actually saw:

  • Recent commits and PR titles (git log --oneline -100) — real change shapes
  • Corrections already recorded in CLAUDE.md — each one is a case that should now pass
  • Past incidents or bugs — each becomes a permanent case

Start with 5–10 good cases rather than 50 vague ones. Coverage grows by accretion.

2. Scaffold the suite

Use the built-in eval runner rather than hand-rolling one:

claude plugin eval init            # interview-driven suite authoring
claude plugin eval init --bare <name>   # blank single-case template

Cases live under evals/ as case.yaml (or prompt.md + graders/*.md). Write each as a prompt plus acceptance checks: tests pass, lint clean, behaviour unchanged, the repo's policy followed.

If the target is a repo's .claude/ configuration rather than a packaged plugin, and the runner cannot resolve it, fall back to a shell runner over claude -p — one non-interactive run per case, asserting the same checks — and say clearly that this is the fallback.

3. Run and record

claude plugin eval . --threshold 0.8

Record the current pass rate as the baseline. Add evals/results/ to .gitignore — run output is not source. A case that already fails is a finding, not a broken eval — either the config needs fixing or the case is wrong. Decide which, explicitly.

4. Wire into CI

Write .github/workflows/sdlc-evals.yml from ${CLAUDE_PLUGIN_ROOT}/templates/ci/sdlc-evals.yml:

  • Runs on pull requests touching CLAUDE.md, .claude/**, evals/**
  • Runs on a schedule (weekly is enough to start)
  • Fails below the threshold, so config changes are gated on it

Evals cost money to run. Say so, keep --runs low in CI, and mention --max-cost-usd for a hard ceiling.

Read the full file on GitHub · 56 lines

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. yesterday First seen · 56 lines · 27 tokens per session scan A 17f432b32080

Subscribe to this mod's changes

evals is a command published in the GitHub repository artemnovichkov/skills (36 stars, last pushed yesterday), licensed MIT. It adds 27 tokens to every session and 603 once invoked, about $0.0001 per session on Opus 5. 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 2026-09-07.