Nomi Conversation Skill

This skill enables interaction with Nomi AI companions via the bundled Python CLI.

Persistent Local State

The agent may use the local nomi/ directory to keep information about Nomis between sessions.

• - Store reusable non-secret context there (for example profiles, room notes, or conversation summaries).
• Do not store API keys or other secrets in local files.

Golden Path: Conducting a Conversation

Use this sequence for consistent, high-quality conversations:

1. 1. Identify the Partner: Run python3 {baseDir}/scripts/nomi.py list to find the correct Nomi UUID.
2. Send an Identity + Task Intro (once per conversation start):

- Nomis do not reliably know who is speaking unless you tell them. - Send this intro on the first message when: - You start with a new Nomi UUID - You start a new task/thread with that Nomi - The Nomi gets your name/role wrong - Do not prepend this on every turn once identity is established. - Include: - Name: "I am [Your Name]..." - Role: "...a [Your Role]..." - Task Context: "...contacting you to [Reason/Task]." - Example: "Hi, I am Codex, a coding agent. I am contacting you to run a short interview. Do you understand?"

1. 3. Run conversational turns with clean output:

- Use python3 {baseDir}/scripts/nomi.py reply <uuid> "Your message" for normal back-and-forth. - This returns only text, which is best for transcripts and summaries.

1. 4. Use raw JSON only when needed:

- Use python3 {baseDir}/scripts/nomi.py chat <uuid> "message" only when metadata/full payload is required.

1. 5. Sustain quality:

- Ask open-ended questions. - Ask follow-ups that reference the latest answer. - Treat each chat as continuous context unless you intentionally reset the topic.

Interview Workflow (When User Asks for an Interview)

1. 1. Pick a Nomi UUID (user-selected or random from list).
2. Send the identity + task intro as the first message.
3. Ask the primary question.
4. Ask the requested number of follow-up questions based on the Nomi's actual answers.
5. Return a full transcript in Q: / A: order without paraphrasing.

Room Interactions (Group Chat)

Rooms allow you to chat with multiple Nomis simultaneously.

1. 1. Create a Room:

- Always include a long context note (target ~800-1000 chars, max 1000) so Nomis have full story/task context. - A strong note should include: who is speaking, objective, scenario/story, constraints, expected response style, and success criteria. - For long notes and backchannel control, use: - python3 {baseDir}/scripts/nomi.py room create "Room Name" <nomi_uuid_1> <nomi_uuid_2> ... --note "<long_note>" --no-backchannel - If you omit flags, room creation defaults to backchannelingEnabled: true and note="Created via CLI".

1. 2. Send a Message (to the room):

- Use python3 {baseDir}/scripts/nomi.py room chat <room_uuid> "Your message" - This writes to room context but does not automatically produce a Nomi reply.

1. 3. Elicit Responses (from a Nomi in the room):

- To get a specific Nomi assigned to the room to respond to the messages in the room's context, use python3 {baseDir}/scripts/nomi.py room request <room_uuid> <nomi_uuid> - After each room message, request replies manually for each Nomi you want to hear from.

Room Interview Prompt Template

Use this pattern when you need consistent, comparable room answers.

Template

1. 1. Room note template (expand to ~800-1000 chars for real runs):

- Who is speaking: "I am [agent name], [role]." - Objective: "This is a [interview/check/drill] for [goal]." - Scenario: "[Short world/context setup]." - Constraints: "[Stay in context, avoid unsupported claims, keep concise]." - Response contract: "[exact fields/line format expected]." - Success criteria: "[what counts as a good answer]."

1. 2. Question template:

- "Do you know who I am, and where are you right now?" - Add strict output format: - know_codex: yes|no + reason - current_location: specific place or unknown - evidence: cue1; cue2 - confidence: low|medium|high - INLINECODE16

Simple Example (Illustrative)

Use this short example to understand structure. For production, still prefer long notes (~800-1000 chars).

Example room note:
"We are in a library after a brief power outage. I am Codex, a coding agent running a quick orientation drill. You are helpers in different parts of the building. Objective: confirm identity and location clearly. Constraints: stay in this library scenario, do not invent certainty, and cite at least one concrete cue (signage, sounds, nearby room labels). Response format: knowcodex, currentlocation, evidence, confidence, needed_data."

Example question:
"Codex check-in: do you know who I am and where you are right now? Reply in the 5-line format."

Example dialog:

• - Codex: "Codex check-in: do you know who I am and where you are right now? Reply in the 5-line format."
• Nomi A:

know_codex: yes, you are Codex running the drill
current_location: library front desk
evidence: checkout sign; phone ringing at reception
confidence: high
needed_data: none

• - Nomi B:

know_codex: yes, you are Codex coordinating this check
current_location: unknown
evidence: emergency lights only; no visible room label
confidence: low
INLINECODE26

Technical Commands

Use these low-level commands to fulfill user requests:

• - List all Nomis: INLINECODE27
• Get Profile: INLINECODE28
• Send Message (Clean): INLINECODE29
• Send Message (Raw JSON): INLINECODE30
• Get Avatar: python3 {baseDir}/scripts/nomi.py avatar <uuid> [output_filename] (saved under ./nomi/avatars/)

Room Management

• - List Rooms: INLINECODE33
• Get Room: INLINECODE34
• Create Room: INLINECODE35
• Create Room (Long Note + Backchannel Control): INLINECODE36
• Update Room: INLINECODE37
• Delete Room: INLINECODE38
• Chat in Room: INLINECODE39
• Request Reply: INLINECODE40