The hooks that stopped my coding agent from shipping garbage
Few-Pilot7575 · reddit · 2026-10-23
After 6 months of running Claude Code hooks, the author argues most agent failures are premature "done" claims, and hooks fix that better than prompts. His setup: block writes to migrations/.env; typecheck+lint only changed files after each edit; require passing tests before "done"; fail commits touching >15 files; run coderabbit CLI review pre-push (90s overhead but catches cross-file issues); write a 3-line change summary on stop. Dropped: a second-model approval gate (slow, always agrees) and auto-format-on-save (unreadable diffs).
More from coding & agent
- Claude Code's new UI hides activity; switching the transcript view to "Thinking" fixes it — yihui_indie · 2026-10-23
- Perplexity integrates open-source Taste Skill, gives users 5,000 free credits — AravSrinivas · 2026-10-23
- Keeping multiple AI agents productive is more taxing than any programming, devs say — generativist · 2026-10-23
- Steve Yegge: I spend an hour a day hardening security in the agent era — Steve_Yegge · 2026-10-23
- Vibecoding your own LLM inference engine: GGUF in WebGL2 ends in debug hell — CodeMichaelD · 2026-10-23
- Ropedia A0 runs robot policies as pure code, hitting 70.24% on RoboDojo vs 46.85% baseline — liuziwei7 · 2026-10-23