eval-scaffold

eval-scaffold is a command for Claude Code from sigistry/marketplace. It costs 24 tokens per session (965 once invoked), scanned A, original, MIT.

A command that creates a runnable evaluation suite for an application's prompts, agents, or other language-model calls. The suite scores outputs so changes can be checked with tests rather than judged from a few examples.

In plain words
What is it for?
Use it to detect the model and test setup, then create evaluations for prompt and agent behaviour using an existing test or evaluation framework.
Why use it?
It replaces subjective spot-checking with repeatable evidence about whether a prompt or agent change improved or harmed results.

Command for Claude Code

Written for Claude Code: argument-hint in frontmatter. Also seen: model in frontmatter.

Part of the llm-app-hardener plugin — 4 skills, 4 commands, 2 agents shipped together

Good fit Use it to detect the model and test setup, then create evaluations for prompt and agent behaviour using an existing test or evaluation framework.

Compare 6 commands from other repositories ↓
Install with agentmods
npx agentmods add commands/sigistry/marketplace/eval-scaffold
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.

Clone the repo
git clone --depth 1 https://github.com/sigistry/marketplace

Made for: Claude Code.

Or install llm-app-hardener, the plugin that ships this one along with the rest of its 4 skills, 4 commands, 2 agents.

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-scaffold

README.md
[![agentmods](https://agentmods.dev/badge/commands/sigistry/marketplace/eval-scaffold.svg)](https://agentmods.dev/commands/sigistry/marketplace/eval-scaffold)
Your own site
<a href="https://agentmods.dev/commands/sigistry/marketplace/eval-scaffold"><img src="https://agentmods.dev/badge/commands/sigistry/marketplace/eval-scaffold.svg" alt="Measured on agentmods" height="20"></a>
Per session 24 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 965 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 $0.00024 $0.00965
Opus 5 $0.00012 $0.00483
Sonnet 5 $0.00005 $0.00193
Haiku 4.5 $0.00002 $0.00097

Measured 4d ago against content hash 6f404549b39e, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-07, from the pricing page.

Security

Grade A, and why

eval-scaffold 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 4d 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/llm-app-hardener/commands/eval-scaffold.md · 45 lines

How it starts

The opening of the file, as written. The whole thing — 45 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Scaffold a runnable eval suite for the prompt, agent, or LLM call site at $ARGUMENTS (default: the app's prompt/agent files under prompts/, agents/, chains/, or the LLM call sites in the current diff). This escapes "vibes-based" development: after this command, a prompt change is graded by tests instead of by reading three outputs. It applies the eval-authoring skill and rides an existing eval framework where one is present rather than reinventing it.

Process

Step 1: Detect the SDK, runner, and any eval framework

Read manifests and imports to fix the stack before writing anything:

Layer Signal
Anthropic SDK anthropic, @anthropic-ai/sdk, client.messages.create
OpenAI SDK openai, OpenAI(, chat.completions.create, responses.create
LangChain langchain, ChatOpenAI, ChatAnthropic, .invoke(
LlamaIndex llama-index, VectorStoreIndex, query_engine
Vercel AI SDK ai, generateText, generateObject, streamText
Pydantic AI pydantic-ai, Agent(, result_type=
Test runner pytest/pyproject.toml, vitest/jest/package.json, go test
Eval framework promptfooconfig.yaml, promptfoo, deepeval, .deepeval, ragas

If promptfoo or DeepEval is already present, generate configs/tests for it (see the skill's references/eval-frameworks.md). If neither is present, generate framework-free tests on the detected unit runner (Vitest/pytest) plus a small LLM-judge helper.

Step 2: Identify what to evaluate

Read the target prompt/agent and name its contract: the inputs it takes, the output shape it promises, and the failure modes worth catching (wrong format, missing field, hallucinated fact, ignored instruction, unsafe content, verbosity). Pull 3–6 representative cases from fixtures, logs, or the prompt's own examples, never invent domain facts.

Step 3: Generate the suite (apply the eval-authoring skill)

Produce four artifacts:

  1. Assertion tests: deterministic checks per case: equals/contains/icontains, regex, is-json + json-schema, not-contains (banned strings), latency/cost budget. Cheap, fast, run every commit.
  2. LLM-as-judge tests: for qualities assertions can't capture (faithfulness, helpfulness, tone). Use a rubric from references/llm-judge-rubrics.md: explicit criteria, a fixed scale, and a pass threshold. Prefer pairwise (new vs. baseline) for regression gates.
  3. Golden-dataset stub: a versioned cases file (jsonl/yaml) with input, expected/assert, and a tags field, plus a short note on how to grow it from production traces.
  4. CI job: a GitHub Actions workflow that runs the suite on PRs and fails the check on a quality regression (assertion failure, or judge score below threshold / below the baseline).

Read the full file on GitHub · 45 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. 4d ago First seen · 45 lines · 24 tokens per session scan A 6f404549b39e

Subscribe to this mod's changes

eval-scaffold is a command published in the GitHub repository sigistry/marketplace (3 stars, last pushed 4d ago), licensed MIT. It adds 24 tokens to every session and 965 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 2026-09-03.