Reading a 47k-line repo: dump groups to files, page through, fan out subagents

For a full-repo comprehension pass, cat related files into one shell call so the harness persists the output to a file, then page through it with offset/limit reads rather than reading dozens of files individually. Run parallel subagents for self-contained subsystems (seeder, test harness, frontend, decision log) and read the core (API, schema, deploy, CLI) yourself, since that is what you will build on. Get line counts of the persisted dumps first so you can request every remaining page in one parallel batch instead of discovering totals one truncation at a time.