Organization repository on GitHub
RuneBench harness for TypeSafe's Jev: bounded action catalog, tick-mode controller and a live dashboard
- Forks
- 1
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.
CommunityJev Agents & MCP
Jevscape
RuneBench fork that connects Jev to RuneScape through an rs-sdk checkout, offering about 50 bounded catalog actions that Jev selects by probability while code executes them.
- Stars
- 12
- Language
- TypeScript
- Published by
- Community
- Author
- Skyvern-AI
- Last verified
- Oct 2026
About this project
A RuneBench fork with a harness that lets TypeSafe's Jev play RuneScape. This repository forks RuneBench, Max Bittker's RuneScape agent benchmark. The harness connects Jev to the game through a separate rs-sdk checkout. See the upstream README for RuneBench documentation. Jev, TypeSafe's "System One," is a non-generative model. It answers typed questions (choice, noul, score) with calibrated probabilities. It writes no text or code and calls no tools. It cannot use RuneBench's OpenCode agent adapter.
Quoted from the project's README. Read the full README.
Works forJev for Classification
Used inJev for Software & DevTools
- Languages
- TypeScript
Highlights
- Jev answers typed choice, noul, and score questions with probabilities and calls no tools.
- Supports burst mode with one action per burst and tick mode that polls during live play.
- Each poll supplies player status, hitpoints, target, last message, and recent events.
- Includes a live stack with dashboard plus local sweep and aggregation scripts.
Watch out
Needs a separate rs-sdk checkout, Bun, and a TypeSafe key; keep one dashboard tab visible because background tabs pause login. Copy reviewed against the repository README on 2026-10-04, not from running it.
Questions about this entry
Short answers drawn from the linked project or page.
No. The catalog carries the game knowledge, so the comparison mostly measures the action space; the README says to always present Jev beside the random-policy baseline run on the same catalog.
The peak window rewards lucky bursts, which favors random play, while Jev gained more total XP through steady play and won tasks needing a multi-step chain like fly fishing.
Chrome pauses background tabs and the game client only logs in while its tab is visible, so the dashboard must have exactly one tab open and focused during login; a second tab takes over the game session. The rest is Bun, a separate rs-sdk checkout kept at ../rs-sdk, and TYPESAFE_API_KEY, with no Docker needed for the local runner. The live stack also carries two rs-sdk patches: the shipped level curve is 2^(level/10) against a live 2^(level/7), and the shipped XP multiplier of 25 only becomes real XP rates with --xprate 1.
Jev answers poll_again_in_ticks with N, 2N, 5, or 10 to set the next poll, the answer timeout equals the poll interval, and a late answer keeps the current action running.
Install bun, keep an rs-sdk checkout at ../rs-sdk, and set TYPESAFE_API_KEY in your environment; no other secret is required and the local runner needs no Docker. Start the live stack with agents/jev/live-stack.sh up ../rs-sdk, then open the dashboard at localhost:7790.
Burst mode runs one action per burst and then Jev picks the next, and the skill sweep uses it. Tick mode asks Jev during live play, with the base poll interval set by --poll-every N. Burst duration is 20 seconds, or 60 at 1x speed, and a game tick at real speed is 400 ms.
About 50 built on rs-sdk, including fishing, chopping, mining, cooking, fighting, buying, selling, dropping, walking, closing dialogs and eating. Jev decides and code executes each action, returning a probability for every offered option. It cannot use RuneBench's OpenCode agent adapter.
More Jev resources like this
20GitHub stars
TypeScript SDK and CLI coding agent where Jev only picks options while your code runs effects, building programs as syntax trees.
Jev Agents & MCP#community#awesomejev#catalog-import
Community
rhighs/jev-codeTypeScript
rhighs/jev-codeTypeScript22GitHub stars
Runnable agent harness that moves pick, score and yes-no loop choices to Jev or Laya in milliseconds while reserving the language model for writing, compiling DECISION.md into executable questions.
Jev Agents & MCP#community#jevbest#catalog-import
Community
zjunlp/JevLoopTypeScript
zjunlp/JevLoopTypeScript35GitHub stars
Diff-aware test selector that sends the git diff to Jev with one question per test, then prints the native filter flags and hands them to vitest, jest, node, bun, Playwright, cargo, or go test.
Jev Agents & MCP#community#jevbest#catalog-import
Community
mizchi/jev-test-filterTypeScript
mizchi/jev-test-filterTypeScript13GitHub stars
Pi extension that classifies each task with Jev over OpenRouter and routes it to a Pareto-optimal model balancing measured quality, cost, and latency under a role policy.
Jev Agents & MCP#community#jevbest#catalog-import
15GitHub stars
Node CLI that lints a file tree against fuzzy natural-language rules in one Markdown file, judging each matched file with a swappable TypeSafe-backed judge.
Jev Agents & MCP#community#jevbest#catalog-import
Community
tyler-dot-earth/patdownTypeScript
tyler-dot-earth/patdownTypeScript106GitHub stars
Decision layer exposing Jev's typed yes-no, choice, and score answers to DeepSeek Harness and MCP hosts, offline by default with every network transmission disclosed per feature.
Jev Agents & MCP#community#jevbest#catalog-import
Community
PerryLink/jevcoreTypeScript
PerryLink/jevcoreTypeScriptFrom the Jev community
Posts from builders shipping with Jev right now.