Local-first visual browser specialist for coding agents. Playwright + MCP + 18 skills + 12 specialist modes.
This repository also configures its own agents. See what visual-browser-agent tells them →
Skill Claude CodeCodex
Audit the current page for accessibility issues by combining structured accessibility-tree checks with screenshot-required visual checks, producing WCAG-referenced findings.
Skill Claude CodeCodex
Study and record animations, transitions, and motion behavior on web pages. Capture timing, easing, and visual quality.
Skill Claude CodeCodex
Enforce default retention periods for run artifacts with content-addressed storage, user keep/export marks, and cleanup that never deletes user-exported files.
Skill Claude CodeCodex
Core browser navigation and interaction skills. Use for basic navigation, clicking, filling forms, and page inspection.
Skill Claude CodeCodex
Study public websites for layout hierarchy, typography, color, and spacing patterns, then produce an evidence-backed inspiration brief that distinguishes observation from implementation.
Skill Claude CodeCodex
Review captured visual evidence (screenshots, recordings) against requirements using vision analysis. Determine pass/fail, confidence, and findings.
Skill Claude CodeCodex
Handle human-in-the-loop workflows for authentication, CAPTCHA, 2FA, sensitive actions, and ambiguous situations.
Skill Claude CodeCodex
Extract lawful public business fields from directories and company sites into normalized, deduplicated lead lists - with human approval required before any outreach.
Skill Claude CodeCodex
Collect documentation, code examples, diagrams, and release notes from public websites into an organized, searchable evidence library with full source attribution.
Skill Claude CodeCodex
Audit responsive behavior across viewports. Check layout, navigation, content reflow, touch targets, and readability.
Skill Claude CodeCodex
Manage artifact retention, cleanup expired files, and enforce storage limits per configuration.
Skill Claude CodeCodex
Monitor authorized social accounts, summarize activity, and draft replies or posts into an approval queue - no public action happens without explicit human approval.
Skill Claude CodeCodex
Review long recordings efficiently by treating them as indexable datasets - chunking windows, progressive review levels, and frame selection instead of replaying whole videos.
Skill Claude CodeCodex
Capture screenshots, recordings, and visual evidence with automatic triggers and manual control.
Skill Claude CodeCodex
Debug visual issues by capturing evidence, comparing states, and analyzing rendering problems.
Skill Claude CodeCodex
Capture baseline screenshots per viewport and route, re-run after changes, compare using pixel/perceptual diff concepts, and report failures with viewport, route, actions, evidence, and confidence.
Skill Claude CodeCodex
Study a public website's rendered design, interactions, responsiveness, and motion using structured inspection and selective screenshots or recordings. Use for design inspiration, UX study, or visual analysis.
Skill Claude CodeCodex
Observe a multi-step web workflow without changing real state - record short clips per stage, extract a step-by-step map, and identify confusing or failure-prone stages.
At most 3 mods per repository are shown here, and a mod shipped inside a plugin is left to that plugin's page — the rest are on their repository pages: