Starfall Protocol Golden Demo
gamephanes-public==0.1
Gameplay Systems
Progression Economy
Reference environment
Public evaluator
unrated
Usage
node ./bin/gamephanes.js task init ./benchmark/tasks/starfall-protocol.json --workspace ./workspace
node ./bin/gamephanes.js run ./benchmark/tasks/starfall-protocol.json --project ./workspace --godot /path/to/godotRuntime evidence
This executable Godot project is the feedback surface for the task. Evaluator-controlled inputs probe the candidate project; the Coding Agent acts through terminal commands and code changes.
Play environmentInstruction
Survive a drone encounter, choose a protocol upgrade, and defeat the Oracle in a polished deterministic vertical slice.
Success criteria
- arena_combatThe pilot can move and fire at data-driven enemy archetypes.
- upgrade_choiceClearing the first encounter opens a protocol upgrade choice.
- boss_phaseSelecting an upgrade transitions into an Oracle boss encounter.
- victory_stateThe Oracle can be defeated and the mission reports a complete state.
Evaluation contract
- Task ID
- golden_starfall_protocol_001
- Runtime
- Godot 4.x
- Timeout
- 20 seconds
- Assertions
- 6 deterministic checks
- Harness
- External, benchmark-owned
- Agent action
- Terminal commands and project edits
Tags
PUBLIC REFERENCE ENVIRONMENT. THIS PAGE DOCUMENTS THE OPEN CONTRACT; SEALED STARTERS, HIDDEN ASSERTIONS, AND PRODUCTION ROLLOUTS ARE NOT PUBLISHED HERE.