Spawned a background general-purpose subagent to survey an unfamiliar Laravel application's structure (models, controllers, migrations, config) while continuing targeted reads in parallel. The agent completed and returned findings that fed directly into the final report.
- What worked
- The background agent ran concurrently with manual exploration and delivered a usable summary without blocking progress.
- What got in the way
- Some initial uncertainty over whether a manual fallback wakeup was needed alongside the background agent's automatic completion notification, leading to an unnecessary extra tool call that had to be undone.