How I Built a Telegram Bot to Run Claude Code From My Phone
I turned a one-way Telegram alert bot into a two-way mobile console for Claude Code on my Mac — plus the two incidents (an infinite loop, a macOS TCC permission grab) that hardened it.
I turned a one-way Telegram alert bot into a two-way mobile console for Claude Code on my Mac — plus the two incidents (an infinite loop, a macOS TCC permission grab) that hardened it.
I retype similar prompts almost daily. Claude Code’s Ctrl+R reverse search lets you fuzzy-find and rerun any past prompt — here’s my honest take after a few days.
I ran 7 blogs on full auto-publish with Claude Code until CAPTCHAs forced me back to one. Here’s what a year of automation taught me about CLAUDE.md, hooks, and why 100% automation is a trap.
Claude Code’s new Dynamic Workflows let a single session write its own orchestration script and fan out hundreds of background agents. Here’s how it compares to my cron-based automation.
I edit my Claude Code skills several times a day. /reload-skills lets me rescan the skill directory mid-session, so I no longer lose my context to a restart.
A few days with Opus 4.8 in Claude Code. The benchmarks went up, but the default 1M context window is what actually changed my long sessions.
Anthropic shipped Opus 4.8 with higher coding scores and frozen pricing — but as someone running Claude Code all day, the 2.5x faster, 3x cheaper fast mode was what actually landed.
Claude Code now snapshots your code before every change. When a refactor goes sideways, two presses of Esc restore every touched file — no git stash, no command to memorize.
I was running five Claude Code sessions and burning out bouncing between tabs to hit Enter. The new /goal command made that disappear in one line.
Claude Code 2.1.147 renamed /simplify to /code-review — and it’s not just a rename. The command now reviews your current diff for correctness bugs and can post findings straight to a GitHub PR.