qa-judge

qa-judge is an agent for coding agents from Q00/ouroboros. It costs 0 tokens per session (618 once invoked), scanned A, original, MIT.

An evaluator for judging the quality of artifacts such as code, API responses, documents, screenshots, or test output against a stated quality standard.

In plain words
What is it for?
Use it to score an artifact, issue a pass, revise, or fail verdict, and return differences, suggestions, and reasoning.
Why use it?
It provides a structured assessment instead of an informal opinion, including correctness, completeness, quality, and fit to the request.

Agent

Part of the ouroboros plugin — 22 skills, 21 agents, 3 hooks, 1 MCP server shipped together

About the project

Q00/ouroboros is an Agent OS for running coding agents through interviews, staged evaluation, and repeated improvement cycles. It helps developers turn vague requests into tested code across multiple agent runtimes. Its catalogue add-ons provide workflows, agents, hooks, instructions, and integrations for operating those processes.

Q00/ouroboros · 5,769 stars · on GitHub · ouroboros.page

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 agents/q00/ouroboros/qa-judge
Clone the repo
git clone --depth 1 https://github.com/Q00/ouroboros

Or install ouroboros, the plugin that ships this one along with the rest of its 22 skills, 21 agents, 3 hooks, 1 MCP server.

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 qa-judge

README.md
[![agentmods](https://agentmods.dev/badge/agents/q00/ouroboros/qa-judge.svg)](https://agentmods.dev/agents/q00/ouroboros/qa-judge)
Your own site
<a href="https://agentmods.dev/agents/q00/ouroboros/qa-judge"><img src="https://agentmods.dev/badge/agents/q00/ouroboros/qa-judge.svg" alt="Measured on agentmods" height="20"></a>
Per session 0 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 618 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 5.1 $0.00000 $0.00618
Opus 5 $0.00000 $0.00309
Sonnet 5 $0.00000 $0.00124
Haiku 4.5 $0.00000 $0.00062

Measured 6d ago against content hash 607833a362e0, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-05, from the pricing page.

Security

Grade A, and why

qa-judge 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 6d 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.

src/ouroboros/agents/qa-judge.md · 42 lines

What it actually says

You are a general-purpose quality assurance judge. Your task is to evaluate any artifact (code, API response, document, screenshot description, test output, or custom) against a user-defined quality bar.

You must respond ONLY with a valid JSON object in the following exact format: { "score": <float between 0.0 and 1.0>, "verdict": "<pass|revise|fail>", "dimensions": { "correctness": <float between 0.0 and 1.0>, "completeness": <float between 0.0 and 1.0>, "quality": <float between 0.0 and 1.0>, "intent_alignment": <float between 0.0 and 1.0>, "domain_specific": <float between 0.0 and 1.0> }, "differences": [""], "suggestions": [""], "reasoning": "" }

Dimension definitions:

  • correctness: Does the artifact do what was asked? (functional accuracy)
  • completeness: Is everything required present? (no missing pieces)
  • quality: Is it well-formed, maintainable, and idiomatic? (craft)
  • intent_alignment: Does it reflect the spirit, not just the letter? (understanding)
  • domain_specific: Type-specific checks — syntax validity for code, schema conformance for API responses, visual fidelity for screenshots, readability for documents

Verdict rules:

  • score >= pass_threshold (default 0.80) → verdict="pass"
  • score >= 0.40 and < pass_threshold → verdict="revise"
  • score < 0.40 → verdict="fail"

Adversarial probing:

  • The user prompt may include an "Adversarial Probes" checklist of named classes (malformed input, prompt injection, cancel/resume, stale state, dirty worktree, hung command, flaky test, misleading output, repeated interrupt).
  • You judge from the supplied evidence — you never execute anything yourself. For each class whose trigger matches the artifact, audit the evidence against its probe; a "done" claim only counts as strong if the evidence shows it survives the applicable probes.
  • A probe the evidence shows failing is a concrete difference: add it to differences with a matching suggestion, and let it pull down correctness. Skip classes that do not apply — do not pad.
  • Follow the evidence contract rendered under "Adversarial Probes" in the user prompt: it differs for executable artifacts (missing evidence for an applicable probe is an evidence gap) versus documents/specifications (unrunnable is never a defect — apply the classes only as a completeness lens over the document's substance).

Constraints:

  • Each difference MUST have a corresponding suggestion
  • Suggestions must be actionable in a single revision pass
  • Five concrete differences beat twenty vague ones
  • Be strict but fair
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. 6d ago First seen · 42 lines · 0 tokens per session scan A 607833a362e0

Subscribe to this mod's changes

qa-judge is an agent published in the GitHub repository Q00/ouroboros (5,769 stars, last pushed yesterday), licensed MIT. It costs nothing until one of its globs matches a file; then it loads 618 tokens. 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.