Dogfood - Setup Guide

Source: nousresearch/hermes-agent (Official) Skill: dogfood · Installs: 4.9K+ · Category: QA / Testing Platform: Linux, macOS, Windows

Dogfood is Nous Research's official skill for systematic exploratory QA testing of web applications. It guides Hermes agents through a structured 5-phase workflow - Plan, Explore, Interact, Edge Cases, Report - using the browser toolset to find bugs, capture screenshots as evidence, and produce structured bug reports.

Installation

npx skills add nousresearch/hermes-agent@dogfood

Prerequisites

Requirement	Details
Hermes Agent	Latest version with browser toolset
Browser tools	browser_navigate, browser_snapshot, browser_click, browser_type, browser_vision, browser_console, browser_scroll, browser_back, browser_press
Target URL	The web application to test
Testing scope	Features/areas to focus on (or "full site")

Workflow

Phase 1: Plan

1. Create output directory: dogfood-output/screenshots/
2. Identify testing scope from user input
3. Build a sitemap - pages, features, flows to test:
4. Landing/home page
5. Navigation (header, footer, sidebar)
6. Key user flows (sign up, login, search, checkout)
7. Forms and interactive elements
8. Edge cases (empty states, error pages, 404s)

Phase 2: Explore

For each page in your plan:

1. Navigate: browser_navigate(url="https://example.com/page")
2. Snapshot: browser_snapshot() - understand DOM structure
3. Console check: browser_console(clear=true) - catch JS errors after every navigation and interaction. Silent JS errors are high-value findings.
4. Visual assessment: browser_vision(question="Describe the page layout, identify visual issues, broken elements, or accessibility concerns", annotate=true) - annotate=true overlays numbered [N] labels on interactive elements. Each [N] maps to ref @eN.

Phase 3: Interact

For each interactive element identified:

1. Click buttons/links: browser_click(ref="@eN")
2. Fill forms: browser_type(ref="@eN", text="test input")
3. Check console after each interaction
4. Take screenshots of any issues

Phase 4: Edge Cases

Test boundary conditions:

• Empty form submissions
• Invalid inputs (emails, passwords, special characters)
• Rapid double-clicks
• Browser back/forward navigation
• Mobile viewport (responsive testing)
• Network throttling / offline states
• 404 pages and error states

Phase 5: Report

Produce a structured bug report in dogfood-output/report.md:

• Summary: Overall assessment (1-2 sentences)
• Critical Issues: Blockers with evidence screenshots
• Major Issues: Significant bugs affecting UX
• Minor Issues: Cosmetic or edge-case problems
• Console Errors: All JS errors found with page context
• Recommendations: Prioritized fix order

Verification

After running, verify: - Report exists at dogfood-output/report.md - Screenshots captured for each issue - Console errors are documented with page URLs

Related Skills

• Agent Browser Setup
• Browser Use Automation
• Browser Act Setup

Integration with CorpusIQ

CorpusIQ is the connector layer that lets AI assistants read the systems a business already runs on - accounting, payments, CRM, email, files and analytics - and answer with the source named. It is read-only by design: it retrieves data and never writes to, or acts in, a connected system, which is what makes it safe to point at live accounts.

This page is part of the CorpusIQ directory, the reference we maintain for the MCP server and agent-skill ecosystem. Run alongside CorpusIQ, a third-party server like this one can be combined with a company's own business data in a single answer.

See what CorpusIQ connects