Agent Functions

A program that controls the agent

15 steps. One text file in. One tested game out. The programmer writes the pipeline. The AI agent does the work. Every step validated. Every decision auditable.

← Back to the story

Step 13: evaluateAndImprove — The core of the pipeline. Each iteration, the AI agent:
1. Starts a local server and loads the game in Playwright
2. Reads window.GAME_STATE bridge to check scene transitions, input, scoring, game loop
3. Takes screenshots of title, gameplay, and game over states
4. Captures all console errors
5. Reads the source code and compares against REQUIREMENTS.md
6. Identifies the single most impactful issue by priority: P0 (crash) > P1 (broken core) > P2 (missing feature) > P3 (polish) > P4 (juice)
7. Creates a git snapshot, then fixes that one issue
8. Appends to EVALUATION_LOG.md for a full history of what was found and fixed
9. Returns gameComplete: true when all requirements are met and no significant issues remain

Max 7 iterations. If the game is complete before that, it exits early to documentation.

Step Reference

#	Function	Description	Output	Next
1	extractGameRequirements	Parse description.txt, detect genre	REQUIREMENTS.md	→ 2
2	createGameDesign	Full Game Design Document	GAME_DESIGN.md	→ 3
3	improveGameDesign	Review and improve design	GAME_DESIGN.md	→ 4
4	evaluateGameDesign	Gate: all requirements covered?	DESIGN_EVALUATION.md	pass→5 \| fail→3 (max 5)
5	createGameArchitecture	Phaser 3 / Matter.js architecture	ARCHITECTURE.md	→ 6
6	improveGameArchitecture	Review and improve architecture	ARCHITECTURE.md	→ 7
7	evaluateGameArchitecture	Gate: all design elements covered?	ARCHITECTURE_EVALUATION.md	pass→8 \| fail→6 (max 5)
8	createImplementationPlan	Step-by-step build plan	PLAN.md	→ 9
9	improveImplementationPlan	Review and improve plan	PLAN.md	→ 10
10	evaluateImplementationPlan	Gate: all architecture covered?	PLAN_EVALUATION.md	pass→11 \| fail→9 (max 5)
11	buildGame	Generate Phaser 3 game files	index.html, game.js, styles.css	→ 12
12	setupTestInfrastructure	Install Playwright + server	package.json, playwright.config.js	→ 13
13	evaluateAndImprove	Play-test, find #1 issue, fix it	EVALUATION_LOG.md, modified files	complete→14 \| else→13 (max 7→14)
14	generateDocumentation	Write README.md	README.md	→ 15
15	generateFinalSummary	Write SUMMARY.md	SUMMARY.md	→ DONE