eval

eval is a skill for Claude Code from GiustoPiedimonte/agentic-engineering-marketplace. It costs 91 tokens per session (810 once invoked), scanned A, original, MIT.

An evaluation skill for judging an AI agent or feature using real failed runs. It groups failures into named problems and builds tests around them.

In plain words
What is it for?
Collecting failure traces, analysing recurring error types, creating evaluation criteria, and running a harness to check regressions.
Why use it?
It replaces guesswork with evidence about what is breaking and whether changes actually improve results.

Skill for Claude Code

Written for Claude Code: $ARGUMENTS substitution. Also seen: mentions subagents.

Part of the agentic-engineering plugin — 6 skills, 5 agents, 2 hooks shipped together

Good fit Collecting failure traces, analysing recurring error types, creating evaluation criteria, and running a harness to check regressions.

Compare 6 skills from other repositories ↓
Install with agentmods
npx agentmods add skills/giustopiedimonte/agentic-engineering-marketplace/eval
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 GiustoPiedimonte/agentic-engineering-marketplace --skill eval
Clone the repo
git clone --depth 1 https://github.com/GiustoPiedimonte/agentic-engineering-marketplace

Made for: Claude Code.

Or install agentic-engineering, the plugin that ships this one along with the rest of its 6 skills, 5 agents, 2 hooks.

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

README.md
[![agentmods](https://agentmods.dev/badge/skills/giustopiedimonte/agentic-engineering-marketplace/eval.svg)](https://agentmods.dev/skills/giustopiedimonte/agentic-engineering-marketplace/eval)
Your own site
<a href="https://agentmods.dev/skills/giustopiedimonte/agentic-engineering-marketplace/eval"><img src="https://agentmods.dev/badge/skills/giustopiedimonte/agentic-engineering-marketplace/eval.svg" alt="Measured on agentmods" height="20"></a>
Per session 91 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 810 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 $0.00091 $0.00810
Opus 5 $0.00046 $0.00405
Sonnet 5 $0.00018 $0.00162
Haiku 4.5 $0.00009 $0.00081

Measured 8d ago against content hash 0af868cf0d85, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-08, from the pricing page.

Security

Grade A, and why

eval 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 8d 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.

plugins/agentic-engineering/skills/eval/SKILL.md · 65 lines

How it starts

The opening of the file, as written. The whole thing — 65 lines — stays where its author put it; the contents beside it link to each section on GitHub.

/eval — make eval the unit of progress

You don't improve what you don't measure, and you don't measure with invented test cases. Evals start from real failures, get coded into failure modes, and become the harness that gates change. $ARGUMENTS is the feature / agent / flow to evaluate (or a pointer to a trace dump).

This skill is offline and diagnostic — it builds and runs the harness over traces/datasets and tells you where and how things break. Its live sibling is /measure (a one-shot read-only verdict on a gated decision); /eval is what defines what "good" even means, and produces the criteria /measure later checks.

Process

  1. Start from real traces, never synthetic. Pull actual failing runs (logs, recorded sessions, the dev DB). Delegate the wide read to the researcher subagent and read-only data pulls to the measurer. Synthetic cases validate the mechanism, not the magnitude — they cannot tell you what really breaks.

  2. Error-analysis (open-code → axial-code). Read a sample of failures and write a free-text note on what went wrong for each. Then cluster those notes into a small set of named failure modes (e.g. wrong-tool-selected, argument-extraction-error, voice/format-drift, hallucinated-fact). Count them — frequency ranks what to fix first. Stop sampling when new traces stop producing new modes (theoretical saturation), and say how many you read.

  3. Localize with a transition-failure matrix (for multi-step / agentic flows). For each step boundary, record where the failure first appears, so a regression is pinned to a stage (tool-selection vs extraction vs formatting) rather than to a vague pass-rate. See references/EVAL_PLAYBOOK.md.

  4. Pick the assertion level per mode — the cheapest that's faithful:

    • Deterministic / component-level (code asserts): exact tool chosen, schema valid, value in range. Prefer these — fast, free, no judge drift.
    • LLM-judge / task-level only where judgment is irreducible (voice, helpfulness, end-to-end success). Align the judge to human labels on a held-out set before trusting it; report its agreement rate.

Read the full file on GitHub · 65 lines

Files

What ships with it

1 file 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. 8d ago First seen · 65 lines · 91 tokens per session scan A 0af868cf0d85

Subscribe to this mod's changes

eval is a skill published in the GitHub repository GiustoPiedimonte/agentic-engineering-marketplace (13 stars, last pushed 1mo ago), licensed MIT. It adds 91 tokens to every session and 810 once invoked, about $0.0005 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 2026-08-30.