Skip to main content

Tools

Things outside Claude Code itself. Verified 2026-09-20.

Session & account

ToolWhat it doesNotes
cc-switchDesktop app to switch between Claude Code providers and accounts★134k. This is what "cswap" in my notes was
CodeIslandLive agent status in the MacBook notch★2.3k. Already wired into my settings.json
vibe-notchSame idea, notifications-focused★2.5k, quieter since April

Skills & discovery

ToolWhat it does
skills.sh + npx skillsThe registry and CLI. Install, update, search skills across GitHub. See skills
vercel-labs/find-skillsA skill that finds skills — npx skills find from inside a session
npx claude-doctorAnalyses your own session history for friction patterns: edit-thrashing, frustration, excessive exploration. Abandoned (v0.0.3, April) — run it once for the report, do not adopt it

claude-doctor's output is the single most useful thing in this list, because it is data about your sessions. Its suggested CLAUDE.md rules are already folded into this kit's.

Token economy

ToolWhat it does
RTKCLI proxy compressing command output before the agent reads it. 60–90% on bash output, 99% on test runs. Full page
headroomSame idea as an MCP server: compresses tool output, logs and RAG chunks
snipCLI proxy, declarative rules, 60–90%

Documents

ToolWhat it does
pdf-inspectorPDF → clean Markdown, and text-vs-scanned classification. No OCR. Skill in this kit
anydocAny document → Markdown. The generalisation of the above
npx -y skills@latest add firecrawl/anydoc

Browser

ToolWhat it does
playwright-mcpDrive a real browser from the agent. The default choice
browser-usePython agent framework for browser automation. Heavier; for building an agent, not for driving one

Design reference

ToolWhat it does
refero.designSearchable library of real product UI. "How do others solve this screen" before generating one
interfaces.dev/cheat-sheetThe rules behind the interface-cheatsheet skill

Benchmarks

ToolWhat it does
deepswe.datacurve.aiCost and performance comparison across coding models

Catalogs