What is claude-referee?

An unofficial Claude Code plugin by ismaildasci that checks a coding agent's 'done' claims and small judgement calls with TypeSafe Jev, and keeps receipts of what was actually verified.

How does it use TypeSafe Jev?

The plugin description says it checks done claims and small judgement calls with TypeSafe Jev, and its topics list jev and llm-as-a-judge. The digest excerpt truncates before documenting the exact questions or endpoints.

What does the done-gate do?

It challenges a claim such as 'All tests pass. Done.' For example it replies that no check has passed since the last edit and asks Claude to run cargo nextest, which reveals 2 failures. The hero description labels active mode as planned.

How do I install claude-referee?

The README has an Install section, but the digest truncates before it, so the exact commands are not available here. It is distributed as a Claude Code plugin from the ismaildasci GitHub repository.

What does the plugin cost to use?

The README has a 'What it costs' section, but the captured excerpt truncates before it. Since the referee calls Jev per check, expect API usage; the repository documents the specifics on its own pages.

What measurements does the repo publish?

The repository carries docs/measurements.md and a 'What's measured so far' section, but the digest excerpt cuts off before any numbers, so this entry cannot report accuracy, latency, or savings results.

Why does the project say 'evidence over eloquence'?

The README's framing: coding agents write well, and 'All tests pass. Done.' is one fluent line that costs nothing to write, while checking it takes a test run. The referee makes the agent show receipts before the claim is believed.

Jev directory

CommunityJev Integrations & Adapters

Claude Referee

An unofficial Claude Code plugin that checks 'done' claims and small judgement calls with TypeSafe Jev and keeps receipts: its done-gate can reject 'all tests pass' and ask for a real check, so evidence beats eloquence.

See highlights Open github.com

Stars: 8
Language: TypeScript
License: MIT
Reading time: 2 min
Published by: Community
Author: ismaildasci
Last verified: Oct 2026

ismaildasci/claude-referee

User repository on GitHub

View on GitHub

Evidence over eloquence. Unofficial Claude Code plugin that checks "done" claims and small judgement calls with TypeSafe Jev, and keeps receipts.

Forks: 2
Last pushed: Updated Oct 2026

Repo stats from the GitHub API, cached Oct 2026.

Tagscommunity github claude-code verification agent

Works forJev for Verification Jev for Decision gating

Used inJev for Software & DevTools

Languages: TypeScript
Platforms: CLI
Capabilities: Open source

Highlights

The done-gate example: Claude says all tests pass and the referee replies that no check has passed since the last edit.
It then asks Claude to run cargo nextest, which shows 2 failures — the README's hero scenario.
An unofficial Claude Code plugin that checks claims and small judgement calls with TypeSafe Jev and keeps receipts.
Topics include ai-agents, claude-code hooks, developer tools, jev, and llm-as-a-judge; a manifesto ships in English and Turkish.
MIT licensed TypeScript: 8 stars, 2 forks, created 2026-09-30, last pushed 2026-10-15.

Watch out

README reviewed 2026-10-17; the digest truncates in the introduction after 'Checking it takes a test run', so install commands, the What it costs section, and docs/measurements.md are not visible here. The active done-gate mode is labeled planned in the hero image description.

Questions about this entry

Short answers drawn from the linked project or page.

Reactions & coverage

Posts, threads, and videos about this entry from around the web.

Reddit: Localization QA with a decision model: meaning, terminology, and placeholders as separate checks.

Hacker News: Pick Verify: a single decision at the warehouse pick station catches confusable picks.

Show HN: Verifying robot picks with a decision modelRead the thread on Hacker News

YouTube: Ray Amjad puts Jev inside an agentic coding loop and looks at what it costs to run.

YouTube: RoboNuggets puts Jev to work inside Claude Code.

From the Jev guides

Original write-ups on what this entry is for.

Decision gates in agent frameworks: what each integration ships

Oct 16, 2026

How Pydantic AI, Mastra, TanStack AI, DSPy, Spring AI, and LiteLLM expose Jev: where the typed decision sits in the loop, what each integration returns, and which big frameworks are only adjacent.

Jev Integrations & Adapters#agents#frameworks#integrations

Read article

From the Jev AI model community

Posts from builders shipping with Jev right now.

Follow @typesafeai