identity

memory-loop: identity

Names live in one local file the SessionStart hook reads:

~/.claude/groundwork/memory-loop/identity.json

Schema:

{
  "userName": "Sam",
  "assistantName": "Iris",
  "chosenBy": "user",
  "status": "set"
}

• chosenBy: "user" (the user named the assistant) or "assistant" (the assistant chose its own name). When it is "assistant", the hook adds "You chose this name yourself." to the injected context.
• status: "set" or "declined".

Set

1. Ask what the user would like to be called.
2. For the assistant's name, offer both paths and let the user pick the path:
   • the user names the assistant → record chosenBy: "user";
   • the assistant chooses its own name and explains why it chose it → record chosenBy: "assistant". Choose something you actually mean — this name persists across sessions.
3. Confirm both names with the user before writing anything.
4. Write the file (create the directory first):

   mkdir -p ~/.claude/groundwork/memory-loop
   

   then write the JSON above with the confirmed values and status: "set".
5. Echo the resulting names back so the user sees exactly what future sessions will inject.

Change

Read the current file, update only the fields the user asked to change, keep the rest as-is, and confirm before writing.

Decline

If the user does not want names, write exactly:

{ "status": "declined" }

and tell them the offer will never appear again — they can opt back in any time by invoking this skill.

Hard rule

Never write identity.json without the user's explicit confirmation of the values. This file speaks as context in every future session; a wrong or unwanted name is a persistent annoyance, not a cosmetic bug.