AI engineering
Using Claude Code in real Android work
I use Claude Code every day, at work and on my own projects. A good result needs more than a good model. It needs rules, hooks and a clear workflow around it.
On this page you can find how I brought it into my team, the tools I built for it, and what I wrote about it.
At work
Bringing it into the team
At Bitpanda I led the adoption of Claude Code in my team. We set up custom rules, sub-agents and skill definitions for our Android codebase. It helped us deliver faster and reduced the manual work in code review.
Outside work I also build small open-source tools for Claude Code. You can find them below.
Open source
Plugin marketplaceTools I built
- Token GuardA plugin that cuts token use in every session.
- Blocks expensive agents when Grep, Glob or Read can do the same job.
- Rewrites noisy commands like git log or Gradle tasks to shorter versions before they run.
- Pure Python, no build step.
- LedgerShows what my Claude Code sessions cost.
- Reads the session files Claude Code already writes, so no API key is needed.
- Terminal dashboard, web dashboard and slash commands.
- Published on npm as @rezaiyan/ledger.
- SkillfetchSyncs skill files from GitHub repos into a project, safely.
- You pick which skills to sync and see the diff before anything is written.
- Your own notes inside synced files stay after updates.
- A scanner blocks prompt injection and leaked credentials.
- claude-kit-v2Gives Claude memory across sessions.
- Saves a summary of every session in SQLite.
- Loads the right context when a new session starts.
- No API calls and no cloud.
- NotifierA desktop notification when Claude is done or needs input.
- Works on macOS and Linux.
- Also works on managed Macs where custom hooks are blocked.