After a revision upgrade, a lite runner from an old checkout fails login with an opaque "Client out of date" and its scripts only report "No game client is attached". - Engine: on a revision mismatch, read the username from the login block (laid out the same across revisions) and record it for 30 minutes; a correct-revision login clears it. The management port serves it at /outdated-client. - Gateway: when an SDK connects to a bot with no live client, ask the engine and answer sdk_error with the update steps. Old SDK checkouts print gateway errors, so this reaches people who haven't pulled yet. - Public /api/version. The lite runner checks it before loading the cache and fails with OutdatedClientError (a LoginError, code 6) naming the fix; the runner and swarm exit 3 instead of retrying. - CLAUDE.md troubleshooting entry for the message. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
17 KiB
RS-Agent Bot Guide
You're here to play the mmo game through the progressive development of botting scripts, starting small then adapting to your desires and ideas. It is strongly recommended to get started and make the first step towards your goals, then researching and learning as you go.
First Time Setup
Create a new bot using the setup script:
Ask the user for a bot name (max 12 chars, alphanumeric). If they skip, use the command without a username to auto-generate a random 9-character name.
# With custom username
bun bots/create-bot.ts {username}
# Auto-generate random username
bun bots/create-bot.ts
# Use local server (sets SERVER=localhost in bot.env)
bun bots/create-bot.ts {username} --local
# Use a custom server
bun bots/create-bot.ts {username} --server=myserver.example.com
This automatically creates:
bots/{username}/bot.env- Credentials with auto-generated passwordbots/{username}/lab_log.md- Session notes templatebots/{username}/script.ts- Ready-to-run starter script
Start a Game Client (scripts only)
Scripts (bun bots/{username}/*.ts) and bun sdk/cli.ts need a running game
client; they never start one. Before running a script, start the headless lite
client in the background and leave it running:
cd server/webclient && bun install && bun src/lite/runner.ts {username}
It logs Gateway connected, registering as '{username}' once it's ready. One
process per bot; it exits non-zero if the game session ends. Without it, a
script exits with No game client is attached. MCP execute_code doesn't need
this step; it opens a browser client itself.
Quick Start
- Install dependencies:
bun install(from project root) - Open project in your AI coding tool — the MCP server will be available automatically
- Control your bot with suggestions.
Tools
| Tool | Description |
|---|---|
execute_code(bot_name, code) |
Run code on a bot. Auto-connects on first use. |
list_bots() |
List connected bots |
disconnect_bot(name) |
Disconnect a bot |
Example
// Just execute - auto-connects on first use
execute_code({
bot_name: "mybot",
code: `
const state = sdk.getState();
console.log('Position:', state.player.worldX, state.player.worldZ);
// Chop trees for 1 minute
const endTime = Date.now() + 60_000;
while (Date.now() < endTime) {
const tree = sdk.findNearbyLoc(/^tree$/i);
if (tree) await bot.chopTree(tree);
}
return sdk.getInventory();
`,
});
See mcp/README.md for detailed API reference.
Code runs in an async context with bot (BotActions) and sdk (BotSDK) available as globals.
Reporting SDK Bugs
When the SDK has a bug or rough edge, first find a workaround, then file a report. One command, no auth, no permission needed:
bun sdk/bug-report.ts "incorrect results from bot.foo(), had to use raw sdk.sendFoo() instead for xyz reason."
Important note: The game itself is extremely well tested and complete, it's not buggy. If you can't figure out how to do something, don't blame the game, blame your assumptions and keep investigating. File a bug report after you figure out what's going on, don't just assume the game is broken because you got confused, and give up. The SDK is a thin layer on top of the game, and the game is the source of truth.
After filing the bug, keep working on your goal!
Session Workflow
This is a persistent character - you don't restart fresh each time. The workflow is:
1. Check World State First
Before writing any script, check where the bot is and what it has:
bun sdk/cli.ts {username}
This shows: position, inventory, skills, nearby NPCs/objects, and more.
Exception: Skip this if you just created the character and know it's at spawn.
Tutorial Check: If the character is in the tutorial area, call await bot.skipTutorial() before running any other scripts. The tutorial blocks normal gameplay.
2. Write Your Script
Edit bots/{username}/script_name.ts with your goal. Keep scripts focused on one task. you may write multiple scripts for different tasks and switch between them.
3. Run the Script
bun bots/{username}/script_name.ts
4. Observe and Iterate
Watch the output. After the script finishes (or fails), check state again:
bun sdk/cli.ts {username}
Record observations in lab_log.md, then improve the script.
Chatting Without Taking Control
To talk in-game (or read chat) without disturbing whatever script currently
owns the bot, use the chat CLI — it connects in observe mode, which can send
chat (and nothing else) and never pre-empts (or gets pre-empted by) a
controller:
bun sdk/chat.ts {username} "meet me at the bank" # send public chat
bun sdk/chat.ts {username} --to {player} "on my way" # send a private message
bun sdk/chat.ts {username} # recent chat
bun sdk/chat.ts {username} --watch # tail live
Public chat is capped at 400 chars per message; sdk.say() auto-splits longer
text into wire-safe chunks.
Private messages (sdk.dm(name, text) / sdk.sendPrivateMessage(name, msg))
reach the target anywhere in the world - no friends-list setup needed - and
show up in getChat() as type 3 (received) / 6 (sent). The server accepts one
social packet (say or PM) per tick; dm() paces its chunks automatically.
Script Duration Guidelines
Start short, extend as you gain confidence:
| Duration | Use When |
|---|---|
| 10s | New script, single actions, untested logic, debugging |
| 30s-1 min | Validated approach, building confidence |
| 5+ min | Proven strategy, grinding runs. USE SPARINGLY |
A failed 5-minute run wastes more time than five 30 second diagnostic runs. Fail fast and start simple.
Look out for "I can't reach" messages - the solution is often to open closed gates or that the item isn't accessible.
Read and grep in the learnings/ and wiki/ folder for tips, skill guides, item and npc locations, and shop information.
SDK API Reference
For the complete method reference, see sdk/API.md (auto-generated from source).
Quick overview:
bot.*- High-level actions that attempt to observe method-specific evidence (chopTree, walkTo, attack, etc.). Check a result when the method returns one.sdk.*- State queries and low-level browser-client dispatches. A successfulsend*result does not prove the game server applied the effect.
bot.* Quick Reference
| Method | What it does |
|---|---|
walkTo(x, z, tolerance?) |
Pathfind to coords, opens doors along the way |
talkTo(target) |
Walk to NPC, start dialog |
interactNpc(target, option?) |
Walk to NPC, interact with any option (e.g. 'trade', 'fish') |
interactLoc(target, option?) |
Walk to loc, interact with any option (e.g. 'mine', 'smelt') |
attack(target) |
Walk to an NPC or player and start combat |
castSpell(target, spell) |
Cast a combat spell on an NPC or player |
pickpocketNpc(target) |
Pickpocket NPC, detects XP gain vs stun |
chopTree(target?) |
Chop tree, wait for logs |
pickupItem(target) |
Pick up ground item |
openDoor(target?) |
Open a door or gate |
openBank() |
Open nearest bank |
depositItem(target, amount?) |
Deposit item to bank |
withdrawItem(slot, amount?) |
Withdraw item from bank |
closeInterface() |
Close any open modal (shop, bank, book, quest scroll) |
closeShop() / closeBank() |
Close the shop / bank interface |
openShop(target?) |
Open shop via shopkeeper NPC |
buyFromShop(target, amount?) |
Buy item from open shop |
sellToShop(target, amount?) |
Sell item to open shop |
equipItem(target) |
Equip from inventory |
unequipItem(target) |
Unequip to inventory |
eatFood(target) |
Eat food, returns HP gained |
useItemOnLoc(item, loc) |
Use inventory item on loc (e.g. fish on range) |
useItemOnNpc(item, npc) |
Use inventory item on NPC |
burnLogs(target?) |
Light logs with tinderbox |
fletchLogs(product?) |
Fletch logs with knife |
craftLeather(product?) |
Craft leather with needle |
smithAtAnvil(product) |
Smith bars at anvil |
dismissBlockingUI() |
Dismiss level-up dialogs (called automatically by all actions) |
navigateDialog(choices) |
Auto-click through dialog options |
skipTutorial() |
Skip the tutorial island |
sdk.* Commonly Used Directly
| Method | What it does |
|---|---|
getState() |
Full world state snapshot |
getSkill(name) / getSkillXp(name) |
Skill info |
getInventory() / findInventoryItem(pattern) |
Inventory queries |
findNearbyNpc(pattern) / findNearbyLoc(pattern) |
Find nearby entities |
findNearbyPlayer(pattern) / getNearbyPlayers() |
Find other players |
findGroundItem(pattern) |
Find ground items |
getDialog() |
Current dialog state |
sendClickDialog(option) |
Click dialog option by its published index |
clickDialogByText(pattern) |
Click dialog option by label — prefer this |
sendClickComponent(id) |
Click interface button |
sendCloseModal() / sendCloseShop() |
Close the open modal / shop |
sendDropItem(slot) |
Drop inventory item |
sendUseItem(slot) |
Use inventory item (bury, etc.) |
sendUseItemOnItem(src, dst) |
Combine two items |
say(text) |
Send chat, auto-chunked past the length cap |
dm(name, text) |
Send a private message, auto-chunked |
getChat(opts?) / getNewChat(opts?) |
Read chat history (500-deep) / only unseen messages |
waitForChat(opts?) |
Wait for a message ({from, matching, timeout}) |
waitForCondition(pred) |
Wait for state predicate |
waitForTicks(n) |
Wait n game ticks |
scanNearbyLocs(radius?) |
Async extended-range loc scan; must be awaited |
See sdk/API.md for exact async signatures, parameter defaults, and return
types.
Code environments
MCP execute_code exposes bot and sdk directly. Standalone
bots/{username}/*.ts files must import runScript and receive
{ bot, sdk } in its callback. Do not paste runner-context expressions into
execute_code.
Only one control connection should own a bot. Use observe mode for read-only
monitoring, avoid overlapping action programs on the same bot, and inspect
sdk.getStateAge() before acting after a disconnect or long pause.
Modals Hide the Inventory
A shop or bank modal replaces the inventory tab, so the server rejects inventory packets sent underneath it as "component not visible" — with no game message. Close the modal before doing anything else:
await bot.closeInterface(); // any modal
await bot.closeShop(); // shop specifically
await sdk.sendCloseModal(); // low-level
Don't click the interface's own "Close Window" component id. It's a client-side close button with no server trigger; the SDK now routes those to a close for you, but the named methods above are what to reach for.
Dialog Options Are Not In Product Order
Skill dialogs (fletching, crafting) publish four buttons per product — Make X, Make 10, Make 5, then one labelled with the product name. Arrow shafts are option 4 of 12, not option 1. Never assume ordering:
const dialog = sdk.getDialog();
console.log(dialog?.options.map((o) => `${o.index}: ${o.text}`));
await sdk.clickDialogByText(/arrow shaft/i); // or bot.fletchLogs('arrow shaft')
Dismiss Level-Up Dialogs
All BotActions methods automatically dismiss blocking UI (level-up dialogs, etc.) before executing. You don't need to call this manually in loops.
// Only needed if using low-level sdk methods directly:
await bot.dismissBlockingUI();
// Or manually check
if (sdk.getState()?.dialog.isOpen) {
await sdk.sendClickDialog(0);
}
Error Handling
const result = await bot.chopTree();
if (!result.success) {
console.log(`Failed: ${result.message}`);
// Handle failure - maybe walk somewhere else
}
Project Structure
bots/
└── {username}/
├── bot.env # Credentials (BOT_USERNAME, PASSWORD, SERVER)
├── lab_log.md # Session notes and observations
└── script.ts # Current script
sdk/
├── index.ts # BotSDK (low-level)
├── actions.ts # BotActions (high-level)
├── cli.ts # CLI for checking state
└── types.ts # Type definitions
learnings/
├── banking.md
└── ...etc
wiki/
├── npcs/
├── items/
├── skills/
└── shops/
Troubleshooting
"No game client is attached" / "No state received" - No client is publishing state for the bot. Start the lite client (cd server/webclient && bun src/lite/runner.ts {username}) or open the bot's browser URL, then rerun.
"Client out of date" / "the server now runs revision N" / lite runner exits with code 3 - The game server was upgraded to a new game revision. Update your checkout (git pull, then cd server/webclient && bun install) and restart the lite runner. Browser clients only need a page reload.
Script stalls - Check for open dialogs (state.dialog.isOpen). Level-ups block everything.
"Can't reach" - Path is blocked. Try walking closer first, or find a different target.
Wrong target - Use more specific regex patterns: /^tree$/i not /tree/i (which matches "tree stump").
Everything succeeds and nothing happens - The two silent failure modes. A
send* returning success only means the client wrote bytes; it is not an ack.
- The client never sent anything. The client pathfinds before every interaction, and when it can't route it returns
cant_reachwithout writing a packet - so the server never replies "I can't reach that!" either. Checkreachableon the target before using it.find*already prefers reachable matches; pass{reachable: true}to getnullinstead of an unreachable one. - The server discarded the op. A refused op (mid-action, stunned, target gone or out of view, bad option) gets
UNSET_MAP_FLAGand nothing else - no message, no error. A blocking modal is worse: the op is accepted and the trigger never runs. Checkstate.opFeedback.opRejectedCountaround a send, and gate loops on an observed effect rather than on a timer.
Start small and build up!