A running index of tools I've designed and shipped with Claude Code — mostly small, single-purpose automations and a few web3/security experiments. Each one is built end-to-end: scraping/parsing, an LLM stage or two where it earns its keep, dedup/state, and a delivery channel (Discord/Telegram/dashboard), then deployed on a schedule.
Open-source repos are linked. Everything here is scrubbed of personal data, credentials, and infrastructure details by design.
Scrapes multiple job boards, dedupes against SQLite, applies config-driven
title/location filters, and posts new listings to Discord or Telegram on a
schedule. Sources, filters, and notifiers are all driven by one YAML file —
drop the repo into an AI coding assistant and it self-onboards from a
CLAUDE.md playbook.
- Stack: Python, BeautifulSoup,
cloudscraper, SQLite - Open source: https://github.com/cprkrn/job-bot
Turns streaming reference playlists into organized, deduplicated Beatport / rekordbox / Discogs crates, plus a "what should I actually buy" ranker. Account-agnostic, no personal data.
- Stack: Python, Spotify/SoundCloud + Beatport/Discogs APIs, pyrekordbox
- Open source: https://github.com/cprkrn/crate-toolkit
Reads the follow graph and surfaces non-mutuals; includes a careful, rate-limited bulk-unfollow tool over a persistent browser session.
- Stack: Python, Playwright
- Open source: https://github.com/cprkrn/ig-unfollower
Watches a ticketing site's embed widget and alerts a chat channel the moment resale inventory appears. Handles bot-protection, posts a daily status heartbeat, and has an experimental Playwright auto-checkout path.
- Stack: Python,
curl_cffi, Discord webhooks, scheduled polling
A personal "daily paper" that assembles todos, lists, a reading queue, and news into one brief and delivers it via chat + a static broadsheet, with a companion bot for quick capture during the day.
- Stack: Python, LLM summarization, Telegram bot, scheduled
Pulls account transactions via Plaid and sends a nightly financial summary over chat, plus polished PDF report generation and a FIRE/savings-rate view.
- Stack: Python, Plaid API, Telegram bot, Docker
For sending one announcement to a large contact list without leaking anyone's number: plans a single relay message per group, dedupes recipients, and keeps the contact list entirely local. Pairs with a small static site.
- Stack: Python, contact-graph planning, Astro + static hosting
Semi-automated search + opt-out across data brokers, search engines, and social platforms; auto-drafts CCPA/GDPR removal requests and tracks status.
- Stack: Python, SQLite
Method writeups for a few larger local-only pipelines (full guides available on request):
- Concert history reconstruction — parse ticket-confirmation emails across 9 ticketing services → cross-service union-find dedupe → a self-contained interactive dashboard of every show you've been to.
- Photo archive toolkit — read-only survey + exact-duplicate finder + an "everywhere I've been" map for a photo library scattered across old drives. macOS, no API keys.
- Apple Health → trend tables — a memory-safe parser for Apple's giant
export.xmlthat produces year-over-year trend tables (VO₂Max, resting HR, HRV, sleep). 100% local. - Mail reconciliation — turn USPS Informed Delivery digests into a mail-watching system that reconciles physical mail against an inbox.
Foundry fork-replays of historical DeFi exploits, rebuilt for learning (e.g. reproducing a well-known reentrancy). Educational, fork-only.
An agent harness that runs a recon → audit → verify → triage pipeline over in-scope contract code, used both for protocol audits and for auto-analyzing public bug-bounty scopes (Code4rena / Sherlock). Fork/testnet-only; any findings are disclosed responsibly.
Audited months of my own Claude Code transcripts (the vast majority of token spend was re-reading context) → built a live context-size meter for the status line, an auto-handoff hook that snapshots a session before it blows past its context budget, and a spend-autopsy script.
Built with Claude Code. Contributions and questions welcome on the open-source repos above.