Claude Code

Verdict. My primary tool, for development and for managing everything I know.

what it is

Anthropic’s agentic coding tool. It runs in the terminal, reads and edits files, runs commands, and extends through slash commands, skills, hooks, subagents and MCP servers. The desktop app was rebuilt in April 2026 around running several sessions side by side, with a sidebar to manage them and a panel for background tasks.

how I use it

I run everything from a single Obsidian vault wired to Claude Code. The vault is the database, a handful of custom slash commands are the app layer, and every loop ends in a file instead of a chat transcript. Raw input gets captured in the vault, Claude processes it, the result goes back into the right folder, and I review it later.

What that covers day to day:

• Vault upkeep, skills, templates and daily logs
• Code generation and project scaffolding
• Research synthesis
• MCP connections to calendar, email, Notion and others

Custom commands I wrote in April 2026:

• /implement-spec reads a spec file and builds it step by step, with checks
• /verify detects the project type and runs lint, types, tests and build
• /commit-push reviews changes, then commits and pushes after I confirm
• /deploy reads the project’s deploy config, then pulls and restarts on the server

The skill frameworks on top are gstack and superpowers.

what I found

From a pass over community skills in April 2026, all installed with npx skills add:

• Kept the official Frontend Design plugin. About 400 tokens, and it forces an aesthetic direction before any frontend code gets written.
• Kept Vercel’s agent skills (npx skills add vercel-labs/agent-skills --yes --global). That’s 100+ accessibility and UX rules, 57 React performance rules ordered by impact, and React composition patterns.
• Skipped Trail of Bits Security. It needs CodeQL and Semgrep installed locally and is built for professional auditors.
• Skipped Webapp Testing. Serializing browser output costs a lot of tokens.
• Skipped superpowers at the time, because of its 22K-token startup cost.

compared to

• Codex, OpenAI’s coding agent. I run it as a second-opinion reviewer through gstack’s /codex.
• opencode, the open-source terminal agent in the same category.