My Easy Claude Code Passive Income AI Automation Setup

A practical “set-and-forget” automation loop can turn Claude Code skills into steady, low-effort side income—by running a headless skill on a timer and wiring the results to actions like bug-bounty monitoring or scheduled email reports. The core method is simple: create a Claude Code skill (a Markdown “skill” file), then trigger it repeatedly from the terminal using a `while true` loop that calls `claude-p` with the skill name and sleeps between runs. The payoff is that the same pattern can support very different workflows without needing a complex server or custom scheduler.

For income, one example centers on the Cali market bug bounty program. The system continuously scans for new bug-bounty opportunities on Cali’s prediction markets, checks and claims relevant items in a structured sequence, and sends an email to the appropriate support address when a qualifying bug is found. The transcript gives a rough payout structure—minor bugs around $25, moderate $50, severe $100, with an additional $10 for pre-listing bugs—and describes the setup as fully autonomous. The creator reports earning roughly $100 to $200 per week at times (with variability), while also noting other projects can make smaller amounts or even lose money.

Technically, the automation is built around three pieces. First, a skill file such as `autoc cali` (stored in a skills directory) contains step-by-step logic: checking for front batches, claiming logs, and grouping by event while following a checklist so runs stay consistent. Second, supporting scripts and data flow handle external inputs—like polling for new markets and writing them to a log (e.g., a `new markets` file). Third, an email automation component (a Python file) integrates into the skill so the workflow can notify the right address when conditions are met.

Once the skill exists, the continuous runner is handled with a terminal loop: `while true; do claude-p /<skill-name>; sleep <seconds>; done` (the transcript uses `sleep 60` as the initial test). After each run finishes, the loop waits and triggers the skill again. Timing is adjustable—switching from 60 seconds to 3,600 seconds (one hour) is suggested for a more useful cadence and for running 24/7.

To demonstrate the same pattern beyond bug bounties, the transcript walks through building an “automail” skill that emails the top five Hacker News posts. The workflow: use Claude Code to create a placeholder skill, then implement steps to fetch the top five posts (including URLs) into a `news.json` file and send an email via a Gmail script using a token JSON for authentication. After updating Claude Code settings to allow the required bash commands, the loop is started again with `claude-p /automail` and a sleep interval. The result is repeated email batches, with the transcript showing that the system can trigger again and send multiple emails per run.

Overall, the key insight is that Claude Code skills plus a simple terminal `while` loop can provide reliable, repeatable automation for monitoring opportunities or distributing curated information—without relying on more fragile third-party scheduling changes—so long as permissions and timing are set correctly.