A detailed guide on setting up Cursor agent hooks to automatically run lint and build checks after each AI agent turn. Covers the `stop` hook lifecycle event, how to return `followup_message` to feed failures back into the agent conversation, and four common pitfalls: Cursor's bundled Node polluting PATH, infinite loops from success follow-ups, stdout corruption from debug output, and incorrect exit codes. Includes reference configuration for `hooks.json` and a bash script skeleton with PATH sanitization, abort detection, JSON-safe output via Python, and loop limits. Also compares `stop` vs `postToolUse` vs `sessionEnd` and discusses security and performance trade-offs.
Table of contents
On this pageBackground and prior artHow Cursor agent hooks workHook events: comparison and mental modelWhy stop enables a self-healing loopImplementing a verification hookPitfalls we hit in productionDebug logging and skipping work on abortReference configurationTrade-offs and alternativesApplications and examplesValidation and measurementSecurity and performance considerationsLimitations and future workTroubleshootingFAQNext stepsSort: