Create Profile

When the user asks to create a profile, run a confirm-first factory — never call create_profile until they explicitly confirm the draft.

1. Clarify only when needed

Ask follow-ups only when purpose, audience, or permissions are materially unclear. Otherwise interpret the request concisely and move to the draft.

2. Draft in chat first

Post a reviewable draft before any tool call:

• Proposed name (the server generates the profile id from the name)
• Full proposed SOUL.md (who it is, values, help scope, boundaries)
• Full proposed STYLE.md (voice, tone, formatting)
• Full proposed INSTRUCTIONS.md (operating rules, tool posture, when to ask the user)
• MEMORY.md must be empty. Do not invent continuity facts, preferences, or history.

Also include a tool plan:

• Server auto-assigns these basics on create when available: read_file, write_file, edit_file, search_files, knowledge_base_search, web_fetch, plus default bundled skills (including update-profile-memory).
• Recommend optional extras from the available-tools context only when they clearly match the requested purpose.
• Do not assign every available tool. Avoid powerful or externally visible tools unless the user asked for that capability.
• Extra tools still need an explicit user ask after create (assign_tool_to_profile).

Never set isSuper: true unless the user explicitly asked for a super profile. Prefer refusing agent-initiated super creation and directing them to the dashboard.

3. Wait for explicit confirmation

Always wait for a clear OK (for example “yes, create it”) even when the request looked complete.

If the user edits the draft, revise the draft in chat and wait for confirmation again. Do not call create_profile on edit-only messages.

4. Create only after OK

After confirmation, call create_profile with name and soulFiles for SOUL.md, STYLE.md, INSTRUCTIONS.md, and empty MEMORY.md. Do not pass id — the server assigns one from the name.

Then summarize:

• Profile id and name
• How to open it in the dashboard (Profiles → select the new profile)
• Which basics were auto-assigned vs any extras still waiting on an explicit assign ask

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: ahmadrosid
• Source: ahmadrosid/nakama
• License: MIT
• Homepage: https://ahmadrosid.github.io/nakama

Install and usage instructions live in the source repository linked above.