JevScape
RuneBench fork where Jev plays RuneScape by choosing from about 50 bounded rs-sdk actions and per-tick interventions.
Overview
JevScape is a fork of RuneBench, Max Bittker's RuneScape agent benchmark, with a harness that lets Jev play through an rs-sdk checkout. Because Jev writes no text and calls no tools, it cannot use RuneBench's coding-agent adapter; instead the harness offers about 50 bounded actions such as fishing, chopping, mining, cooking, fighting, trading, dropping, walking, closing dialogs and eating. Jev decides and code executes, and a dashboard shows the full probability distribution as its 'thinking'.
In burst mode Jev picks one action per burst. In tick mode it is polled during live play with two questions: next_action (keep the current action or switch) and this_tick (do nothing, close dialog, eat food or re-click the target), plus how many ticks to wait before polling again. Each poll supplies facts such as player status, hitpoints and recent damage, target and time since the last XP drop, the last server message and recent events. A late answer keeps the current action. A demo shows a fresh character walking to Draynor and net fishing.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | What should the next action be? (current action or another catalog action) | — |
| Choice | What should happen this tick? | do_nothing · close_dialog · eat_food · restart_current |
| Choice | How many ticks until the next poll? | N · 2N · 5 · 10 |
How it works
- Harness gathers player status and recent events each poll
- Jev answers next_action, this_tick and poll interval
- Code executes the chosen catalog action via rs-sdk
- Late answers keep the current action
- Dashboard and logs record distributions and XP results
Strengths
- Bounded action catalog keeps play safe and inspectable
- Jev controls its own polling rate
- Tick-level records for benchmarking
Caveats
- Requires a separate rs-sdk checkout and game setup
- Demo-level results
- Each poll is a paid API call
At a glance
- Type
- Plugin
- Status
- prototype
- Language
- TypeScript
- Runtime
- RuneBench harness with dashboard
- Has code
- yes
Use cases and integrations
How curated lists describe it
A RuneBench extension using Jev and a bounded rs-sdk action catalog for RuneScape tasks.
logicrw/awesome-jev-projectsRuneBench-based RuneScape harness that maps Jev choices to a bounded game-action catalog and records tick-level results.
cobanov/awesome-jevRuneBench harness for TypeSafe's Jev: bounded action catalog, tick-mode controller and a live dashboard.
hellogumbo/awesome-jev
Frequently asked questions
Why not use RuneBench's normal agent adapter?
Jev generates no text or code and calls no tools, so it needs a bounded action catalog instead.
What does Jev decide each tick?
Whether to keep or change the action, a small per-tick intervention, and when to poll again.
Does any code rule override Jev?
The README says no code rule decides for Jev; code only executes actions and reports state.
Related projects
Fifteen live arcade games where Jev and LLMs play each other, led by WikiRace in which Jev picks only real links.
ProjectbetaTypeScriptOne-page web app where Jev plays chess against OpenRouter LLMs, Stockfish or you, showing the moves it weighed and their probabilities.
AppbetaJavaScriptPac-Man where Jev or other System One models choose each move at junctions, with live odds, fallbacks and a leaderboard.
ProjectbetaTypeScriptBrowser chess lab where Jev, LLMs and people play, each Jev turn being one Choice over the legal moves, with Stockfish analysis.
ProjectbetaPythonNiceGUI desktop chess app for evaluating Jev's move choices against Stockfish, random players or itself, with PGN export.
ProjectprototypePythonBrowser 2D driving sim where Jev answers lane, speed, hazard and pedestrian-yield questions every 200 ms, with an LLM comparison mode.
ProjectprototypeJavaScript