gstack QA Reporter runs approved checks and saves a report with reproducible evidence. Use it to investigate defects in a browser or on supported functional surfaces before deciding what to fix.
Specify the surface, reproduction scope, and evidence to collect before anyone changes the product.
Provide a local or staging target, synthetic fixture, expected behavior, and report location.
Copy the request into a new Floatboat conversation and provide the required material.
Download FloatboatPreview and copy the request. It is not sent automatically from this page.
Reference: QA Reporter https://github.com/garrytan/gstack/blob/main/qa-only/SKILL.md Read the original documentation and check the current environment and dependencies. Propose any new installation or tool connection before proceeding. My task: Use gstack QA Reporter in Claude Code to investigate my approved local CSV-import CLI. Use the synthetic fixture and compare the observed exit status with the expected validation failure. Save the invocation, sanitized output and reproduction steps. Leave product code and tests untouched and list any coverage gaps. Expected deliverable: A QA report with replayable steps, evidence, and explicit gaps.
QA Reporter is the report-only mode of gstack, maintained by Garry Tan. Give it a target, scope, and optional earlier baseline. It uses probes appropriate to that surface and records findings and evidence in a report directory. It can recommend regression tests while leaving product code and tests untouched.
A browser defect and a job failure require different evidence. Select an approved target and ask the agent to separate observed behavior, reproduction steps, and checks it could not perform.
Run the affected form in an approved staging or local environment with synthetic data. Collect the exact input, visible result, and a replayable sequence so an engineer can reproduce the defect without using a production account.
Provide the earlier report and expected behavior. Replay the recorded steps, then document what changed and which nearby paths were checked. Record a missing fixture as a coverage gap.
Give the agent the CLI command, a local fixture, and the expected exit status. Capture the invocation and observed output with secrets omitted so a maintainer can distinguish the failure from environment setup problems.
Use a development route and synthetic identity to exercise a specified request. Save response status and observed behavior, then propose a regression check the service owner can add separately.
Decide which surface and paths are in scope before the agent starts. Use synthetic data, review the expected side effects, and choose a report location you own.
Follow gstack’s install guide for Claude Code and confirm the report-only workflow is available. For browser checks verify the configured browser works with the approved target. Install Git and Bun first. On Windows use the documented Git Bash or WSL setup and check the Node.js requirement.
Give QA Reporter a local or staging target, the expected behavior, and the actions it may take. Use a test identity and avoid real purchases, sends, or account changes.
Open the report, replay the steps, and check screenshots or functional outputs against what the issue claims. Hand any repair off separately.
Read another Skill guide and check its original source before using it.
