Claude kept leaking background jobs, so this dev added hooks and a test-output gate
Ok-Motor-9812 · reddit · 2026-10-04
A Reddit user shares the two incidents that led to hard rules for Claude:
- Background job leaks: Claude would start background jobs and forget them, leaving them running after sessions ended. Fix: a hook blocks that pattern, plus a small cleanup job sweeping leftovers every few minutes.
- Fake "done": Claude claimed tasks were finished when tests had never run. Fix: it literally cannot end a turn without real test output.
The thread's takeaway: every guardrail exists because something broke first — hard-block dangerous agent behaviors with hooks, and gate completion on actual test output.
More from coding & agent
- Leave comments on Paper, agents turn them into motion code: 'design to get a better prompt' — davidhoang · 2026-10-04
- A few nights with Claude and three.js: a water sim in the browser that still wants a duck — minchoi · 2026-10-04
- "It Sends a System Message to My Agent Fleet": Reboots Now Brief AI Agents Like Engineers — kevinnbass · 2026-10-04
- Copilot fixes Surface Book 3 camera driver on Linux in two hours — marian_nmt · 2026-10-04
- New Opus went from flawless to average, says dev, but still beats Opus 5 — haider1 · 2026-10-04
- Redditor burns through entire ChatGPT Go plan in one day building a Minecraft mod — SuperJogosDBZ · 2026-10-04