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.
npx agentmods add instructions/dvarahq/evals4j/claude-mdgit clone --depth 1 https://github.com/dvarahq/evals4jWrote 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.
[](https://agentmods.dev/instructions/dvarahq/evals4j/claude-md)<a href="https://agentmods.dev/instructions/dvarahq/evals4j/claude-md"><img src="https://agentmods.dev/badge/instructions/dvarahq/evals4j/claude-md.svg" alt="Measured on agentmods" height="20"></a>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.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.01095 | $0.01095 |
| Opus 5 | $0.00548 | $0.00548 |
| Sonnet 5 | $0.00219 | $0.00219 |
| Haiku 4.5 | $0.00110 | $0.00110 |
Grade A, and why
evals4j CLAUDE.md 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 5d 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.
How it starts
The opening of the file, as written. The whole thing — 84 lines — stays where its author put it; the contents beside it link to each section on GitHub.
evals4j
A Java port of LangChain's OpenEvals (MIT, reviewed at
commit 43fd6af, v0.2.0) for Spring AI and LangChain4j. Published to Maven Central under
com.dvarahq.oss; 0.4.0 is live.
Layout
Maven multi-module, Java 17 baseline (both Spring AI 2.0.0 and LangChain4j 1.18.1 ship
major-version-61 bytecode), built on JDK 21. Root package com.dvarahq.oss.evals4j.
| module | notes |
|---|---|
evals4j-core |
every evaluator, prompt catalog, SPI. No AI-framework dependency — keep it that way |
evals4j-springai / evals4j-langchain4j |
thin adapters implementing JudgeModel |
evals4j-spring-boot-starter |
autoconfiguration; Spring AI wins when both are present |
evals4j-sandbox |
Docker + local SandboxRunner |
evals4j-junit5 |
assertions, EvalReport, @EvalSuite |
evals4j-examples |
runnable examples. Built by CI, never published |
evals4j-bom |
dependency management |
The whole framework boundary is four small interfaces in core/spi: JudgeModel (the only one that
matters — schema in, JSON out), EmbeddingProvider, EvalTracer, SandboxRunner. Everything else
funnels through ScorerRunner, which is what keeps result shapes consistent across evaluators.
Build
./mvnw verify # unit tests, fully offline, no API key
OPENAI_API_KEY=... ./mvnw -Pit verify # plus end-to-end tests against a real model
Unit tests must stay offline — FakeJudgeModel (in core's test-jar, also published) scripts judge
responses. Sandbox tests skip themselves when no Docker daemon is reachable.
Parity discipline
This is a port, and the value is in matching upstream, not in improving on it.
- The 33 prompts are byte-identical to upstream and live as classpath resources, not Java text
blocks, so no escaping rule can alter them.
PromptParityTestchecks all 33 against SHA-256 checksums inevals4j-core/src/test/resources/parity/prompt-checksums.txt. Do not reformat them. - Trajectory and JSON-match behaviour is pinned to upstream's own test fixtures (45 cases transcribed from the Python suite). If a change makes those fail, the change is wrong unless upstream changed too.
- Deliberate deviations — including two upstream bugs fixed rather than reproduced — are documented in PARITY.md. Add to it rather than diverging silently.
- The one upstream capability not reproduced is LangSmith export; there is no Java SDK.
EvalTraceris the seam if that ever 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.
- 5d ago First seen · 84 lines · 1,095 tokens per session scan A 1f8fc2dfc2a9
evals4j CLAUDE.md is an instructions file published in the GitHub repository dvarahq/evals4j (5 stars, last pushed 7d ago), licensed MIT. It adds 1,095 tokens to every session, about $0.0055 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 2026-08-31.
Other instructions, from other repositories
questdb CLAUDE.md
Claude Code instructions for questdb/questdb, covering claude.md, project overview, coding guidelines, tests and questdb's sql dialect.
liteflow CLAUDE.md
Instructions for dromara/liteflow, covering claude.md, 概述, 命令, 构建 and 构建整个项目（模块集由 jdk 版本自动选择 —— 见下方说明）.
BitcoinAddressFinder CLAUDE.md
Claude Code instructions for bernardladenthin/BitcoinAddressFinder, covering claude.md — bitcoinaddressfinder, project overview, build system, common commands and compile only.
imgui-java CLAUDE.md
Claude Code instructions for SpaiR/imgui-java: This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.
pcl copilot-instructions.md
Copilot instructions for personamanagmentlayer/pcl, covering project overview, pcl bootstrap system, core architecture principles, 1. compiler pipeline and 2. type system.
nano CLAUDE.md
Claude Code instructions for NanoNative/nano, covering claude.md, project overview, build & development commands, maven commands and graalvm native image.