proofrag

An evaluation toolkit for retrieval-augmented generation (RAG) and other language-model applications. RAG systems search a document collection before generating an answer.

In plain words
What is it for?
Create an evaluation dataset from your documents, measure correctness and whether answers are grounded in the source material, produce reports, and gate continuous integration on quality.
Why use it?
It replaces ad hoc testing with repeatable checks for answer quality and helps detect regressions after changes to prompts, models, or search settings.

Skill for Claude CodeCodex

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.

agentmods
npx agentmods add skills/unshdee/proofrag/proofrag
Any agent
npx skills add unshDee/proofrag --skill proofrag
Clone the repo
git clone --depth 1 https://github.com/unshDee/proofrag

Made for: Claude Code, Codex.

Per session 95 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 1,882 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
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 $0.00095 $0.01882
Opus 5 $0.00048 $0.00941
Sonnet 5 $0.00019 $0.00376
Haiku 4.5 $0.00010 $0.00188

Measured 2d ago against content hash fb0bcc9dedc8, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

proofrag 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 2d 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.

skills/proofrag/SKILL.md · 150 lines

How it starts

The opening of the file, as written. The whole thing — 150 lines — stays where its author put it; the contents beside it link to each section on GitHub.

proofrag

Turn "did my change make the RAG better or worse?" into one reproducible command. You (the agent) wire the user's app to the kit; the kit does dataset generation, judging, and reporting.

When to use

  • User changed a prompt, model, chunker, embedder, or retriever and wants to know if quality moved.
  • User has docs/a knowledge base but no evaluation set.
  • User wants a hallucination/groundedness number, or a CI gate on answer quality.

Install the engine

This skill drives the proofrag CLI. Make sure it's on PATH (install once), or run it ad-hoc with uvx:

uv tool install "proofrag[anthropic]"     # or: pipx install "proofrag[anthropic]"
# no install needed: uvx "proofrag[anthropic]" demo

Use [openai] instead of [anthropic] for an OpenAI-compatible/local backend. Credentials: ANTHROPIC_API_KEY (default, cheap Haiku judge) or OPENAI_API_KEY (OPENAI_BASE_URL for local/Ollama). No key? proofrag demo renders a sample scorecard. If both keys exist, auto-detection chooses Anthropic; use PROOFRAG_PROVIDER=openai to override. Proofrag does not auto-load .env, so source it first.

The loop

  1. Inspect and generate from the user's corpus.

    proofrag corpus ./docs
    

    Use --include, --exclude, and --no-gitignore when the docs tree is noisy. PDF loading needs the proofrag[pdf] extra; HTML is supported by default.

    Then generate:

    proofrag generate --corpus ./docs --out goldenset.jsonl --n 20
    

    Produces JSONL: {id, question, gold_answer, gold_contexts[], difficulty, sources[]} with tiers single_doc / multi_doc / unanswerable, plus context_metadata for each gold context. Commit this file — it is versioned.

  2. Validate the golden set before committing it.

    proofrag validate --goldenset goldenset.jsonl --corpus ./docs --out validation.json
    

    This checks the JSONL contract, duplicate ids/questions, answerable cases without gold contexts, unanswerable cases that still cite context, source coverage, and a stable fingerprint. It exits non-zero on hard errors; add --strict to fail on warnings too.

Read the full file on GitHub · 150 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. 2d ago First seen · 150 lines · 95 tokens per session scan A fb0bcc9dedc8

Subscribe to this mod's changes

proofrag is a skill published in the GitHub repository unshDee/proofrag (2 stars, last pushed 23d ago), licensed MIT. It adds 95 tokens to every session and 1,882 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 2026-08-31.

Related

Other skills, from other repositories

generate-rag-dataset

Generate a synthetic evaluation dataset from your RAG knowledge base. Creates diverse Q&A pairs with expected answers and relevant context, ready for LangWatch experiments and platform import. Use when you need test data for your RAG pipeline.

langwatch/langwatch · 51 tokens

rag-eval

Use when the user has a RAG (Retrieval-Augmented Generation) system and wants to evaluate its quality — separating retrieval issues from generation issues. Also use when the user mentions RAG evaluation, faithfulness checking, hallucination detection in RAG, retrieval quality, chunking optimization, or "is my RAG…

agentscope-ai/OpenJudge · 86 tokens

autorag-setup

Configure AutoRAG for first use or repair its single-agent model, approved document roots, retrieval indexes, datasource skills, and health checks without exposing credentials.

Marker-Inc-Korea/AutoRAG · 36 tokens

autorag

Use an already configured AutoRAG librarian agent to search, summarize, compare, and answer questions from local document collections. Use autorag-setup for configuration or indexing changes.

Marker-Inc-Korea/AutoRAG · 38 tokens

rag-evaluator

Evaluate RAG pipeline quality across faithfulness, relevance, and hallucination metrics. Use when user asks to test, benchmark, or improve a RAG system, or when RAG outputs look wrong.

chandrudp29/skillhub · 44 tokens

ai-engineering-toolkit

6 production-ready AI engineering workflows: prompt evaluation (8-dimension scoring), context budget planning, RAG pipeline design, agent security audit (65-point checklist), eval harness building, and product sense coaching.

iradoweck/antigravity-awesome-skills · 47 tokens