blog

Field notes from the browser-bashing trenches

Case studies and playbooks for QA engineers and SDETs — every command runnable, every tradeoff stated honestly.

comparison

40 Lines of Page Objects vs. One English Sentence

An SDET ports a 40-line Selenium page-object login test to one BrowserBash sentence, then weighs determinism, speed, and LLM cost honestly.

read →
llm

A/B Test Claude vs Llama on One Suite, One Flag

Swap --model openrouter/<vendor>/<model> to A/B test Claude vs Llama on the same markdown suite — no test changes. Includes OPENROUTER_API_KEY setup.

read →
ci

CI Verdicts Without Log Parsing: Exit Codes 0-3

Wire BrowserBash into GitHub Actions: exit codes 0/1/2/3 as the verdict, NDJSON artifacts for debugging, and flaky-test detection via duration_ms.

read →
case-study

Exploratory Testing With an AI Agent, No Scripts

Release week without throwaway scripts: a QA engineer runs one-shot browserbash objectives for exploratory checks and stores evidence with store-as.

read →
security

Keep Passwords Out of Test Logs with secret:true

A fintech QA team passes credentials via {{vars}} with secret:true — auditors grep 90 days of logs and find only *****. Files, precedence, guarantees.

read →
ci

Local Chrome to Cloud Grids by Changing One Flag

How an agency QA team runs one markdown suite on local Chrome in dev and on LambdaTest or BrowserStack grids in CI by changing only --provider.

read →
case-study

Markdown Tests Your PM Reads, Reviews, and Edits

Tests as living documentation: a product manager reviews and edits *_test.md files in pull requests, with an @import shared-steps library.

read →
case-study

Smoke Tests in Plain English: Markdown Beats Selectors

How a 4-person QA team replaced 38 brittle selector scripts with 12 markdown *_test.md files — and cut smoke-suite maintenance from hours to minutes.

read →
agents

Wire Your AI Agent to a Real Browser With NDJSON

A platform team lets its AI coding agent call browserbash --agent and consume NDJSON: stable schema, step events, exit-code verdicts, bash and jq.

read →
llm

Zero-Budget Browser Automation with Local Ollama

Run a natural-language browser suite on local Ollama for $0 — auto-resolution order, model-size tradeoffs, and OLLAMA_BASE_URL for vLLM or LM Studio.

read →