Skip to content

ulpi Skills - Setup Guide

Source: ulpi-io/skills via skills.sh - 10.7K combined installs across 60 indexed listings; first seen Oct 9, 2026 (extended-query probe) GitHub: ulpi-io/skills (5 stars, NO LICENSE FILE; pushed 2026-08-31; layout: one top-level directory per skill, such as frontend-design-ui-ux/SKILL.md) Category: Web Tooling / SEO / Dev Workflow Quality Tier: 🟡 Beta - 60-skill toolkit; NO LICENSE file; browse tooling notable; mixed sampled verdicts - see Security

ulpi-io publishes a 60-skill toolkit for AI coding agents, and its identity comes from the browse family. browse is a persistent headless Chromium daemon with 76+ commands and ref-based interaction; its README claims 13x fewer tokens than @playwright/mcp because every action returns a one-line result instead of dumping page context. The set reaches 10.7K combined installs across 60 indexed listings, led by frontend-design-ui-ux at 2,393 installs, and covers stealth browsing (browse-stealth via the camoufox runtime, with Turnstile and DataDome bypass plus proxy rotation), SEO/AEO/GEO/QA audits driven through the same browser, and codemap for code search and architecture analysis.

The rest is a complete dev-workflow loop: cost-estimate (299 installs) for costing a repo or branch, code-simplify (185) and branch-review-before-pr (166) for review, bugfix (175) for red-green fixes, find-bugs (168) for a security pass on a diff, git skills (commit, create-pr, git-merge-expert, plus a worktree variant at 159), learnings-propagation skills, and framework references for Laravel, Filament, Next.js, Node.js, NestJS, Rust, and Docker. The launch-* family prepares Product Hunt, Hacker News, X, and LinkedIn launches as composable skills. One caveat up front: the repository carries NO LICENSE FILE.

Installation

Prerequisites: Node.js for the npx skills CLI. The browse skills need extra global installs: npm install -g @ulpi/browse for the browser daemon, and npm install camoufox-js && npx camoufox-js fetch for browse-stealth.

# Install the full suite
npx skills add https://github.com/ulpi-io/skills

# Or a single skill (recommended for this toolkit)
npx skills add https://github.com/ulpi-io/skills --skill browse

What It Provides

Install counts from the Oct 10, 2026 skills.sh snapshot:

Skill Installs Use For
frontend-design-ui-ux 2,393 Locked design language and UX/UI spec in .ulpi/design: anti-slop rules, browse-driven inspiration, design-system routing, a11y; delegates the build (no code)
laravel-filament 431 Filament v5 admin panels: resources, schemas, tables, actions, widgets, and the v3-to-v5 migration
cost-estimate 299 Estimate the dev cost of a repo, branch, or commit
browse 272 Headless browser CLI: 76+ commands, ref-based interaction, 13x fewer tokens than @playwright/mcp
code-simplify 185 Review code for reuse, quality, and efficiency
git-merge-expert 182 Merge branches, resolve conflicts, and roll back
nextjs 181 Installed-version-aware Next.js App Router reference: i18n, trusted data access, public-page contracts
bugfix 175 Red-green bug fixing: reproducer, root cause, minimal fix, regression tests
map-project 174 Generate CLAUDE.md from a codebase scan
plan-founder-review 169 Technical founder review of a plan before execution
find-bugs 168 Security audit and bug finding on a branch diff
plan-to-task-list-with-dag 167 Decompose features into parallel-ready task DAGs
branch-review-before-pr 166 Structural review before a PR: race conditions, trust boundaries
codemap 164 Code search and architecture analysis: hybrid vector/BM25, dependency graphs, PageRank
update-claude-learnings 162 Extract behavioral learnings into CLAUDE.md
update-agent-learnings 160 Propagate learnings to agent files
map-project-monorepo 160 Per-package CLAUDE.md for monorepos
nodejs 159 Node.js/Bun backend reference: TS-first, pino, Zod, async, queues, testing
git-merge-expert-worktree 159 Isolated merges in git worktrees
commit 158 Smart conventional commits, pre-commit checks, secret scanning
update-skill-learnings 155 Propagate learnings into skill files
create-pr 155 Auto-generate PR title and body, push, create via gh
start 151 Session init: discover skills, select agent persona

The remaining 37 indexed listings range from 51 to 149 installs. The browse family continues just below the table line: browse-qa (146), browse-stealth (125), browse-seo (109), browse-aeo (109), browse-geo (109), and browse-config (106), while the launch-* set sits at 51 to 54 installs each.

Why This Matters for Hermes Agents

Most agent browser integrations are context sinks: Playwright MCP dumps roughly 16K tokens per action, while browse returns a one-liner (the README's example session runs about 12K tokens over 10 steps versus 146K, 13x less). For a Hermes agent that reads pages, fills forms, or scrapes structured data, that ratio decides whether a long session stays usable. The audit skills point the same browser at SEO, AEO, and GEO surfaces, which is where discoverability is increasingly decided. codemap gives the agent hybrid vector/BM25 search plus dependency graphs and PageRank when a repo is too large to read directly. The dev-workflow cluster covers a full loop: cost the work, plan it as a task DAG, review the branch against race conditions and trust boundaries, fix with red-green discipline, and merge cleanly. Two caveats belong up front: the repository has no license file, and the sampled verdicts are mixed (a Snyk Warn on frontend-design-ui-ux, a Socket Warn on laravel). browse-stealth exists to get around bot detection, which can conflict with site terms of service; treat it as a tool for sites you are allowed to automate.

Usage

You say What happens
"Audit this landing page for SEO and fix the meta tags" browse-seo runs the on-page audit: meta tags, headings, schema, Core Web Vitals, mobile rendering
"Check how my brand shows up in AI search" browse-aeo audits answer-engine surfaces and browse-geo monitors generative-engine visibility
"Open this site and pull the product listing data" browse drives the persistent Chromium daemon with @ref-based interaction and one-line outputs
"Review this branch before I open a PR" branch-review-before-pr checks race conditions and trust boundaries; find-bugs adds a security pass on the diff
"Turn this feature plan into parallel work" plan-to-task-list-with-dag decomposes the feature into a task DAG
"Fix this bug with a proper reproducer" bugfix runs the red-green loop: reproduce, root-cause, minimal fix, regression tests
"Merge this branch and resolve the conflicts" git-merge-expert merges, resolves conflicts, and can roll back; the worktree variant does it in isolation

Verification

# Confirm the suite installed (paths depend on your agent)
npx skills list | grep -i ulpi

# Review the top skill straight from GitHub before installing (returns 200)
curl -s https://raw.githubusercontent.com/ulpi-io/skills/main/frontend-design-ui-ux/SKILL.md | head -20

# Smoke-test the browse daemon after the global install
browse goto https://example.com

If browse returns a one-line navigation result, the daemon is up; if not, reinstall @ulpi/browse globally. For browse-stealth, confirm camoufox fetched its runtime during install.

Security

skills.sh verdicts for sampled skills (verified Oct 10, 2026) - re-check the security pages on skills.sh before production use:

Skill Gen Agent Trust Hub Socket Snyk
frontend-design-ui-ux Pass Pass Warn
laravel Pass Warn Pass
vue - - -

Two of the three sampled skills carry a Warn on one engine each: frontend-design-ui-ux on Snyk, laravel on Socket. vue returned no verdict data. Only three skills were sampled out of 60; the browse family in particular was not sampled, so audit it directly for your use case.

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

Limitations

  • NO LICENSE file in the repository: there is no explicit license grant; confirm terms with the publisher before commercial use.
  • Mixed security profile: frontend-design-ui-ux has a Snyk Warn and laravel a Socket Warn in the sampled set; most of the 60 skills are unsampled.
  • Small repository: 5 GitHub stars, last pushed Aug 31, 2026; independent and young.
  • The browse skills need separate global installs (@ulpi/browse daemon; camoufox for stealth), so setup is heavier than a plain skill.
  • Stealth browsing (anti-detection, proxy rotation) can conflict with site terms of service; use it only where automation is permitted.
  • Snapshot data, verified Oct 10, 2026: 10.7K combined installs across 60 indexed listings; 5 GitHub stars; NO LICENSE FILE; last pushed Aug 31, 2026. Counts drift over time.

Pro Tips

  • Install per skill: 60 listings is a lot of context; use npx skills add https://github.com/ulpi-io/skills --skill <name> for just what you need.
  • The browse family shares one daemon: install @ulpi/browse once, then use browse goto and browse snapshot -i for @ref-based interaction that returns token-light output.
  • The 76+ browse commands include 150+ device emulation profiles, cookie import from real browsers, command recording, and cursor-interactive detection.
  • Use codemap for architecture questions on large repos: hybrid vector/BM25 search, dependency graphs, PageRank ranking.
  • The launch-* family composes: launch-copy feeds the platform skills (Product Hunt, Hacker News, X, LinkedIn) and launch-analytics handles UTM/GA4 attribution.

← Skills Catalog | Skills Marketplace → Powered by CorpusIQ

AI answers on your data. See where every number comes from.

Connect your tools and ask in ChatGPT, Claude, or Perplexity. Enterprise-grade answers in minutes. From $29.95/month.

Start free, no credit card
Contact Terms Privacy Pricing