RunThru

Product Introduction

1. Definition: RunThru is a GitHub application and AI-powered visual QA automation tool designed for software development teams. Operating in the category of developer tooling and continuous integration (CI) verification, it functions by interacting directly with pull request (PR) preview deployments in a real browser environment.
2. Core Value Proposition: RunThru exists to bridge the visibility gap between code diffs and functional reality. By leveraging autonomous AI agents to test features, record narrated video walkthroughs, and utilize secondary grading models to output a structured verdict (judge.json), RunThru provides software teams with automated visual proof that code changes work before merging, targeting keywords like "pull request video walkthrough", "AI QA automation", and "automated PR verification".

Main Features

1. Autonomous Browser Interaction & Script Generation: When a user comments /runthru on a GitHub pull request, an AI agent parses the code diff, identifies the modified UI flows, opens the deploy preview (from providers like Vercel, Netlify, Render, or Cloudflare Pages) in a real browser, executes the necessary interaction steps, and dynamically writes and voices a corresponding script.
2. Dual-Model Verification and Grading: RunThru decouples execution from evaluation. While an exploratory model (such as Claude Opus) navigates the app and captures video, a separate model (such as Gemini Pro) acts as an independent QA judge. It analyzes the captured screen frames against the generated claims, outputs a structured judge.json containing a score, a pass/fail boolean, and timestamped issue logs with suggested code fixes.
3. Multi-Channel Delivery & Integration: The tool automatically posts results directly into the GitHub PR comment thread, updating dynamically upon subsequent code pushes. Additionally, it integrates with Slack for channel-wide or targeted notifications, generates standalone shareable web pages complete with chapter markers and transcripts, and exports various aspect ratios (16:9, 1:1, 9:16) for cross-functional accessibility.

Problems Solved

1. Pain Point: Reviewing pull requests solely through text-based code diffs leads to merged bugs, untested UI regressions, and a lack of visual verification—especially with the high volume of code generated by AI coding assistants like Claude Code, Cursor, and GitHub Copilot.
2. Target Audience: Frontend Developers, Full-Stack Engineers, Product Managers (PMs), QA Leads, and Engineering Managers who need rapid verification of user-facing changes without manually spinning up preview branches or checking out local test environments.
3. Use Cases: Automatically testing UI components and multi-step user flows on staging environments; generating asynchronous, narrated demos for PMs and designers to review features without pulling code; and running automated sanity checks on AI-generated PRs prior to merging into production branches.

Unique Advantages

1. Strengths & Limitations (Pros & Cons):
   • Pros: Eliminates manual context-switching for reviewers; operates asynchronously within standard GitHub workflows; requires no code installation in the host repository; provides objective, dual-model QA grading with timestamped bug flags; makes engineering changes immediately accessible to non-technical stakeholders via shareable video links.
   • Cons: Relies on the availability and stability of external preview deployments; limited by the complexity of authentication gates (requiring manual test_data or vercel_bypass configuration for protected environments); incurs per-video pricing tiers beyond the free allowance.
2. Key Alternatives & Differentiation:
   • Traditional Test Automation (Cypress, Playwright): While robust, these frameworks require manual script writing, continuous maintenance of test selectors, and active suite management. RunThru dynamically adapts to each PR's diff without requiring test maintenance.
   • AI Code Reviewers (CodeRabbit, PR-Agent): These tools analyze static code diffs for syntax, security, and logic. RunThru goes a step further by actually executing the code inside a live browser environment to evaluate functional UI behavior.
   • Manual QA / Loom Recordings: Traditional manual workflows require developers or QA engineers to manually record and narrate walkthroughs. RunThru automates this entirely via a simple GitHub PR comment command.

Frequently Asked Questions (FAQ)

1. What triggers a RunThru execution on a pull request? Commenting /runthru on any active GitHub pull request initiates the pipeline, provided the repository has the RunThru GitHub App installed and a valid deploy preview is detected or supplied.
2. Does RunThru require changes to my application codebase? No. RunThru operates externally via GitHub webhooks, requiring zero SDK integrations, test files, or dependencies added to your repository (though an optional .runthru.yml configuration file can be used to set custom test credentials or prompt instructions).
3. How does RunThru handle applications behind authentication or login walls? You can pass demo credentials and starting navigation prompts within a .runthru.yml configuration file under test_data, utilize Vercel deployment protection bypass secrets for private repositories, or leverage self-hosted runners to reuse active signed-in sessions.
4. Who can watch the generated RunThru videos? Anyone with the shareable video link can watch the walkthrough, view chapters, read transcripts, and inspect the QA verdict without needing a GitHub account, paid seat, or code-level access.