Documentation Index
Fetch the complete documentation index at: https://quashbugs.com/docs/llms.txt
Use this file to discover all available pages before exploring further.
Quash works differently from traditional test automation tools. There are no scripts to write, no selectors to define, no frameworks to configure. Instead, you describe what you want to test in plain English and two AI agents handle the rest — one generates the tests, the other executes them on your device.
This sounds simple, and it is to use. But getting consistently good output — tests that are specific to your app, that cover edge cases you didn’t think of, that reference your actual screen names and API contracts — requires understanding how the system thinks.
Everything in Quash comes down to one idea: context determines quality. The more the agents know about your app, the better the output. That knowledge builds across three layers — what you configure, what happens in a session, and what the agent learns through experience. Each layer compounds on the others, and none of them ever resets.
The four pages below explain the system. Read them once before your first serious Test Studio session. You will use Quash more effectively from day one, and you will know exactly what to adjust when the output is not what you expected.
Megumi & Mahoraga
Quash has two agents with distinct roles. Megumi is the test generation agent — it reads your context and produces structured test cases. Mahoraga is the execution agent — it runs on your device and carries out the instructions. They never operate at the same time. Understanding where one ends and the other begins is the foundation for using Quash effectively.
Read this if: you are new to Quash and want to understand what happens when you write a prompt versus when you run a test.
→ Megumi & Mahoraga
Memory
Memory is how Quash builds and maintains context across sessions. It operates in three layers: Apps context (what you configure), Recipe Memory (what happens inside a session), and Guidance (what the agent learns over time). The difference between generic tests and tests that feel like they were written by someone who knows your product comes down to how much memory the agent has to draw on.
Read this if: you want to understand why your first session produces different output from your tenth, or why attaching a PRD before prompting makes such a visible difference.
→ Memory
Guidance
Guidance is the knowledge the Abgumi agent builds for itself through repeated testing. You do not write it — it accumulates automatically as Mahoraga runs tests on your app. It starts empty and grows every session, mapping screens, learning flows, and recording edge cases. It is also editable, which matters when your app changes between releases.
Read this if: you have been using Quash for a few sessions and want to understand what the agent has learned, how to review it, and when to correct it.
→ Guidance
Tokens & /compact
Every recipe session operates within a token budget — a measure of how much context Recipe is actively holding at once. Long sessions with many prompts, large documents, and dozens of generated tests can approach the limit. When they do, the agent’s output quality starts to degrade. The /compact command compresses the conversation history so the session can continue without losing what matters.
Read this if: you are running long or complex Test Studio sessions, the token counter is climbing, or the agent’s responses are starting to feel less precise than they were earlier in the conversation.
→ Tokens & /compact