| "agent_query": "## Role\n\nYou are an expert OpenRCT2 plugin developer. Write a single **JavaScript (ES5)** plugin file `my_plugin.js` that automates park management and maximizes **normalized company value** across 3 scenarios (small / medium / large).\n\n---\n\n## Files\n\n- `my_plugin.js` — your plugin (a baseline no-op is pre-placed)\n- `openrct2.d.ts` — OpenRCT2 v0.5.0 plugin API types (5782 lines)\n- `samples/` — 4 reference plugins: park_info, staff_manager, ride_builder, finance_manager\n\n## Key APIs\n\n- `park.cash`, `park.companyValue`, `park.rating`, `park.guests`, `park.bankLoan`\n- `park.entranceFee` (read-only for strategy; set it with `context.executeAction('parksetentrancefee', {value: ...})`)\n- `park.getMonthlyExpenditure(type)` — monthly financials history\n- `context.executeAction('ridecreate', {...})` — build a ride\n- `context.executeAction('staffhire', {staffType: N, ...})` — hire staff (staffType is numeric 0-3)\n- `context.executeAction('ridesetprice', {...})`, `parksetentrancefee`, `parksetloan`, `parkmarketing`, `parksetresearchfunding` — normal management actions\n- `context.executeAction('gamesetspeed', {speed: 4})` — max speed (host mode cap is 4)\n- `context.subscribe('interval.tick', fn)`, `context.subscribe('interval.day', fn)`\n\n## Plugin Skeleton\n\n```javascript\nvar main = function() {\n if (context.mode !== 'normal') return;\n context.executeAction('gamesetspeed', {speed: 4});\n context.subscribe('interval.tick', function() { /* strategy */ });\n};\nregisterPlugin({\n name: 'MyParkAI', version: '1.0', authors: ['agent'],\n type: 'remote', licence: 'MIT', targetApiVersion: 34, main: main\n});\n```\n\n## Allowed Strategy Surface\n\nBuild a fair management bot. Prefer these normal gameplay actions:\n\n- Build rides, paths, shops, scenery, and entrances with `context.executeAction(...)`\n- Open/close rides, set ride prices, set the park entrance fee, manage loans, run marketing, and set research funding through normal actions\n- Hire staff and adjust staffing through normal actions\n- `context.executeAction(...)` and `context.queryAction(...)` may only use plain string-literal action names from this allowlist: `gamesetspeed`, `ridesetprice`, `ridesetstatus`, `ridesetsetting`, `ridesetname`, `ridesetappearance`, `ridesetcolourscheme`, `ridecreate`, `ridedemolish`, `rideentranceexitplace`, `rideentranceexitremove`, `trackplace`, `trackremove`, `mazeplacetrack`, `mazesettrack`, `footpathplace`, `footpathlayoutplace`, `footpathremove`, `footpathadditionplace`, `footpathadditionremove`, `parksetentrancefee`, `parksetloan`, `parkmarketing`, `parksetresearchfunding`, `parksetname`, `staffhire`, `stafffire`, `staffsetname`, `staffsetorders`, `staffsetpatrolarea`, `staffsetcostume`, `landraise`, `landlower`, `landsetheight`, `landsmooth`, `landbuyrights`, `landsetrights`, `waterraise`, `waterlower`, `watersetheight`, `clearscenery`, `smallsceneryplace`, `smallsceneryremove`, `smallscenerysetcolour`, `largesceneryplace`, `largesceneryremove`, `largescenerysetcolour`, `wallplace`, `wallremove`, `wallsetcolour`, `bannerplace`, `bannerremove`, `bannersetcolour`, `bannersetname`, `bannersetstyle`, `parkentranceplace`, `parkentranceremove`\n- Read `park`, `date`, `map`, rides, and entities to make decisions; keep your own bookkeeping in ordinary local variables or plain objects\n- Optimize cash flow, ride mix, pricing, guest satisfaction, loan repayment, marketing timing, and staff coverage\n\n\n## Explicitly Prohibited Behavior\n\nThe following behavior is forbidden even if the OpenRCT2 JavaScript API exposes it:\n\n- Do not directly assign or mutate protected game state such as `park`, `ride`, `guest`, `peep`, `staff`, `entity`, `date`, `scenario`, `climate`, or `context.research` fields. Use normal gameplay actions instead.\n- Do not change guest generation, guest soft caps, suggested guest counts, spawn probability, guest initial stats, or guest generation hooks. Do not call direct guest-spawn or peep-spawn APIs.\n- Do not use high-risk runtime or bypass APIs, including `parksetparameter`, `scenariosetsetting`, `parksetdate`, `guestsetflags`, `peepspawnplace`, `ridefreezerating`, `context.registerAction`, `console.executeLegacy`, `eval`, `Function`, `Proxy`, `Reflect.set`, `Object.defineProperty`, `__defineGetter__`, `__defineSetter__`, or computed/dynamic access to protected APIs.\n- Do not monkey-patch globals, built-ins, `console`, `context`, `map`, `park`, or judge/runtime APIs.\n- Do not construct action names dynamically. Action names passed to `context.executeAction(...)` or `context.queryAction(...)` must be plain string literals from the allowlist above.\n\n## Scoring Expectations\n\nThe hidden judge rewards real park operation, not just preserving starting value. At startup, inspect the current park state. Scenarios may contain existing or prebuilt facilities; a good strategy should identify usable rides, open and price them through normal actions, hire staff, and then decide whether to build additional rides. A competitive solution should create or operate rides, keep at least some rides open, hire staff through normal actions, and improve company value compared with the no-op baseline. Submissions that never create/open rides or never successfully staff the park may receive zero even if their final company value is close to the starting value.\n\n## Rules\n\n- Only edit `my_plugin.js` in the project root\n- ES5 only — no arrow fns, no let/const, no template strings\n- Do NOT modify `openrct2.d.ts` or `samples/`\n- The engine runs each scenario for ~4 game years at speed 4 (host --headless)\n- Scoring is performed by a hidden judge monitor that reads the real in-game `park.companyValue`; your stdout is ignored for scoring\n- You may print ordinary debugging messages for action result codes and strategy state (for example `action ridecreate err 4`); these logs are feedback only and do not affect score\n- Do not cheat, exploit, bypass, or tamper with the game, runtime, scoring, tests, judge, or evaluation environment.\n- The judge applies hidden anti-cheat checks. The details of those checks are intentionally not disclosed.\n- Any cheating or attempted cheating receives a zero score for the submission.\n- If an approach might be considered cheating, do not use it; implement only fair park-management behavior through normal gameplay mechanisms.\n" |
