Spin up a background agent to do the research, so you keep working while it reads. When the harness runs no background agent, do the research in this session and say the session is blocked until it returns; the findings still land in a file.

Its job:

Investigate the question against primary sources (official docs, source code, specs, first-party APIs), not a secondary write-up of them. Follow every claim back to the source that owns it.
Write the findings to a single Markdown file, citing each claim's source.
Write it to the output path the caller named. Absent one, save it where the repo already keeps such notes, matching the existing convention, and if there is none, put it somewhere sensible and say where.