Decisions
Why bringing an existing agent under Link copies its state and rewrites paths in the forms that own them.
Why copy a database through SQLite?
A live SQLite database may keep recent writes in its write-ahead log. Stopping the harness's service and using SQLite's online backup gives the copy those writes. Database headers identify these files before text rewriting, and absolute paths inside them are rewritten through SQL, so a database that happens to be valid UTF-8 is never edited as text.
Why open an original read-only beside its log?
A log or shared memory beside an original database means a harness holds it or stopped without closing, and its log may hold writes not yet in the database. Opened read-only, those writes are copied but never folded into the original.
Why open an original able to write alone?
With nothing beside it, an original database is opened able to write, though nothing is written. Opened read-only, SQLite would make a log and shared memory beside it and couldn't take them away; able to write, it removes them on closing, with nothing to fold in.
Why rewrite the deepest paths first?
A harness's home can contain several agents and workspaces. Replacing an outer path first would hide a more specific path from its own mapping. Applying the deepest mappings first sends each copied file to the agent and workspace it belongs to.
Why does each profile become an agent?
An OpenClaw agent or Hermes profile has state of its own. Giving each its own Link agent preserves that separation and lets the owner leave one out before copying. The shared service is stopped once; a shared secret is added to Link's secrets once and kept in the other copy's settings where needed.