Lessons building a session watcher for 7 AI coding agents: hooks beat process detection
Weak_Cherry_4878 · reddit · 2026-10-18
The author open-sourced a local panel (MIT) watching sessions across 7 AI coding agents. Key lessons: process/CPU detection can't tell which session needs you — agent-reported lifecycle events via hooks are what matter; every agent exposes a different locator (tmux pane id, Codex thread UUID), so normalizing into one status set is the real job; deep links like codex://threads/<uuid> beat window-title matching; external tools should POST to a local HTTP API on loopback instead of parsing stdout; and non-focus-stealing notifications are the underrated feature.
More from coding & agent
- Agent memory's blind spot: lessons learned in one task silently become rules for the next — blaizedsouza · 2026-10-18
- Multi-agent coding is 'the greatest slot machine casinos could never match' — theteknosaur · 2026-10-18
- AI makes it too easy to build every feature — a spec keeps shipping disciplined — audrow · 2026-10-18
- Dev uses Claude to mine 1,270 Apple Notes from 10 years into a private archive of recurring ideas — inexistences · 2026-10-18
- Dev releases LIT Power Tools: a 44-tool local MCP server plus LM Studio memory plugin — LunderIT · 2026-10-18
- Agent self-audit cuts monthly bookkeeping from 3-4 hours to under an hour — RachelVT42 · 2026-10-18