Plan and run game tests with multi-agent automation

POC that plans and executes tests for web puzzle/math games using LangChain agents + Playwright — HTML/JSON reports and artifacts per run.

Designed agent loop + Playwright runner

01 Hook
Manual QA on puzzle games is slow — agents can plan cases and drive the browser.
02 Context
Proof-of-concept: LangChain + Playwright against a local LLM (Ollama), with a small API to plan and run top test candidates.
03 Problem
Game UIs need exploratory testing that scripts alone under-specify. Humans shouldn't hand-write every path before a smoke run.
04 Insight
Split planning from execution: generate a plan, then run the highest-value cases and capture artifacts for review.
05 Exploration
Wired Chromium via Playwright, local llama via Ollama, and endpoints for plan / run / status polling.
06 Decision
Kept scope as a POC — report HTML/JSON + artifacts folders instead of a full CI product.
07 Solution
Operators hit Generate Plan → Run Top 10; results land in reports/ and artifacts/ with curl-friendly APIs.
08 Impact
Demonstrates agentic testing judgment: plan → prioritize → execute → report. Open source POC for hiring conversations.

Impact