Skillget

claude-plugins

mcp
Guvenlik Denetimi
Basarisiz
Health Gecti
  • License — License: MIT
  • Description — Repository has a description
  • Active repo — Last push 0 days ago
  • Community trust — 10 GitHub stars
Code Basarisiz
  • process.env — Environment variable access in plugins/brw/build.sh
  • child_process — Shell command execution capability in plugins/brw/skills/brw/scripts/brw.js
  • exec() — Shell command execution in plugins/brw/skills/brw/scripts/brw.js
  • process.env — Environment variable access in plugins/brw/skills/brw/scripts/brw.js
  • fs module — File system access in plugins/brw/skills/brw/scripts/brw.js
  • network request — Outbound network request in plugins/brw/src/cli/http.ts
  • fs module — File system access in plugins/brw/src/cli/main.ts
  • process.env — Environment variable access in plugins/brw/src/cli/proxy-launcher.ts
  • execSync — Synchronous shell command execution in plugins/brw/src/proxy/chrome.ts
Permissions Gecti
  • Permissions — No dangerous permissions requested
Purpose
This project is a collection of plugins for Claude Code, with the primary tool being "brw" for full browser automation via the Chrome DevTools Protocol. It allows users to navigate, click, type, take screenshots, and intercept network requests directly from their terminal.

Security Assessment
Overall Risk: High. This tool requires extensive system permissions to function. It heavily relies on executing shell commands and running synchronous shell executions (`exec`, `execSync`) to control the host machine and browser. It accesses environment variables and the file system, likely to save screenshots, GIFs, and browsing data. Additionally, it makes outbound network requests to facilitate web browsing. While no hardcoded secrets were found, giving an automated agent the ability to execute arbitrary commands and control a browser inherently exposes highly sensitive data. One plugin even automates tax filing, which requires handling personal financial documents.

Quality Assessment
Overall Quality: Medium. The project is actively maintained, with repository activity as recent as today. It is properly licensed under the permissive MIT license. However, community trust and adoption are quite low, currently sitting at only 10 GitHub stars. The description frames the repository as "random" plugins, which slightly detracts from its professional reliability.

Verdict
Use with caution: While the code functions as intended and is active, the inherent security risks of granting automated shell command execution and browser control to an AI require strict oversight and isolated environments.
SUMMARY

Shrivu's random Claude Code plugins.

README.md

claude-plugins

Blog
GitHub
X
LinkedIn
Chat

Shrivu's random Claude Code plugins.

Plugins

brw

Full browser automation for Claude Code via Chrome DevTools Protocol. Navigate, click, type, screenshot, record GIFs, intercept network requests, and more — all from your terminal.

Why brw? Claude for Chrome is a black box that requires a subscription. Playwright MCP and Chrome DevTools MCP servers don't handle highly parallel agent workflows well — they weren't designed for multiple agents sharing one browser concurrently. brw is a lightweight proxy built for agent-first usage: stateless CLI, per-tab mutexes, and JSON output that agents can parse directly.

/plugin marketplace add sshh12/claude-plugins
/plugin install brw@shrivu-plugins

Then use /brw to start browsing. The proxy auto-starts on first command and stays running. Key capabilities:

  • Navigate & interact: click, type, scroll, drag, keyboard shortcuts
  • Read pages: screenshots, DOM extraction, console logs, network traffic
  • Record: GIF recordings of multi-step workflows
  • Multi-tab: open, switch, and manage browser tabs
  • Security: configurable URL blocking, protocol restrictions, cookie scoping

freetaxusa

Tax filing automation for FreeTaxUSA.com. Guides you through filing US federal and state taxes by extracting your tax documents, researching current tax rules, and automating browser-based form filling via brw.

/plugin marketplace add sshh12/claude-plugins
/plugin install freetaxusa@shrivu-plugins

Then use /freetaxusa to start. The skill walks through 5 phases:

  • Setup: Explains risks, legal disclaimers, loads browser automation, gets consent
  • Discovery: Extracts text and images from your PDFs (W-2s, 1099s, etc.), builds a consolidated summary, front-loads all questions
  • Research: Fetches current-year tax brackets and rules, runs a life-change questionnaire, flags items that may need a CPA
  • Filing: Fills FreeTaxUSA forms section by section with a 4-source verification table (expected, entered, page read, screenshot)
  • Review: Generates a personalized HTML tax breakdown report, walks through a review checklist, creates a data purge reminder

Not a tax professional. Not tax advice. The user reviews and submits their return themselves — the skill never auto-submits.

whatsup

WhatsApp messaging for Claude Code via the Baileys WebSocket client. Send and receive messages, react, share media/locations/polls, search chat history, and long-poll for incoming messages — all restricted to an allowlist of approved contacts.

/plugin marketplace add sshh12/claude-plugins
/plugin install whatsup@shrivu-plugins

Then use /whatsup to start. The proxy auto-starts on first command. Key capabilities:

  • Messaging: send text, media, locations, polls, contact cards, reactions
  • Monitoring: long-poll for incoming messages, read chat history, search messages
  • Profile: set status text, update display name and picture
  • Security: allowlist-only sends, rate limiting, audit logging, untrusted content tagging
  • Architecture: same CLI + proxy pattern as brw — stateless CLI, persistent Fastify daemon, auto-start/stop

diy-mcp-connector

Turn any web app into a set of tools that Claude can call directly — no browser automation, no screenshots, no copy-pasting between tabs. Instead of Claude navigating a UI like a human (screenshot, click, wait, screenshot again), your connector talks to the app's API and returns clean, structured data. One tool call replaces 4-6 browser steps with ~80% fewer tokens and ~10x faster responses.

/plugin marketplace add sshh12/claude-plugins
/plugin install diy-mcp-connector@shrivu-plugins

Then tell Claude which app you want to connect — "I want to connect my recipe app so I can search recipes and pull ingredient lists into Claude." The skill walks through 9 stages: capturing API traffic, designing tools around your workflows, security review, building the server, testing, and connecting it to your MCP client. You don't write the code — Claude does.

  • Uses your existing login — sign in through Chrome once and the connector remembers your session
  • API discovery — auto-detects browser tools you have (Claude for Chrome, brw, Playwright) for live exploration, or works with manual HAR file captures
  • Read-only by default — Claude can look up your data but can't accidentally change anything
  • Works everywhere — Claude Code, Claude Desktop/Cowork, OpenClaw, or any MCP client

cc-essentials

Meta-skills for working with Claude Code itself. Currently ships one skill — build-skill — that walks you through building your own project skill end-to-end: framing the problem, mapping data sources, writing SKILL.md, testing on real tasks, and committing it to your repo.

/plugin marketplace add sshh12/claude-plugins
/plugin install cc-essentials@shrivu-plugins

Then just ask — "help me build a skill for weekly status reports", "I keep re-explaining how to triage these issues, can you turn it into a skill?", or "how do skills work?" The build-skill skill triggers automatically and walks you through it.

  • Problem-first — starts with the painful workflow you actually have, not a hypothetical
  • Data mapping — figures out which built-in tools, MCP servers, and CLIs the skill needs before writing a line of SKILL.md
  • Description tuning — tests the trigger description with near-miss prompts so the skill fires when you want and stays quiet when you don't
  • Real-task testing — runs the skill on 2-3 actual examples and compares against your manual baseline before declaring it done
  • Project-scoped — skills land in .claude/skills/<skill-name>/ and get committed alongside the code, so the rest of your team gets them on the next pull

Yorumlar (0)

Sonuc bulunamadi