Create Group Prompt

Create Group Prompt Files

Step 0: Language selection

Before proceeding with any other steps in this skill, ask the user which language to continue in using AskUserQuestion. Keep this initial prompt in English because the preferred language is not yet known.

• Question: Which language should I continue in?
• Options: English, 日本語 (Japanese)

Use the selected language for all subsequent user-facing messages and for every further AskUserQuestion prompt in this skill. Do not translate code, file paths, shell commands, or file contents.

Create additional prompt files for a group. These files are automatically loaded into the agent's systemPrompt at session start (via agent-runner-claudecode).

UX Note: Use AskUserQuestion for all user-facing questions.

Step 1: Select channel and group

First, list available groups:

find deploy/{ASSISTANT_NAME}/groups/ -type d -mindepth 2 -maxdepth 2 | sort

AskUserQuestion: Which channel and group?

Show the existing groups as options, e.g.:

• slack/main
• discord/main
• Other (specify channel and group name)

If the user picks "Other", ask for the channel name and group name separately.

Step 2: Check existing prompt files

List what's already in the selected group directory:

ls -la deploy/{ASSISTANT_NAME}/groups/{channel}/{group}/

Show the user what files already exist. Explain that CLAUDE.md is loaded by the SDK automatically, and any other *.md files are loaded into systemPrompt.append.

Step 3: Choose which files to create

AskUserQuestion: Which prompt files do you want to create?

Present these common options (multi-select):

• IDENTITY.md — Agent personality, name, speech style, language
• SOUL.md — Mission, core values, behavioral principles
• INSTRUCTIONS.md — Tool usage rules, security rules, structured output specs
• AGENTS.md — Available tools inventory and usage guidelines
• Custom — Specify a custom filename (must end in .md)

Skip any files that already exist (inform the user). If the user wants to overwrite, confirm first.

Step 4: Gather content for each file

For each selected file, ask the user what to include:

IDENTITY.md

AskUserQuestion: Tell me about the agent's identity:

• Name / nickname
• Character / personality (e.g., "zundamon-style character", professional assistant, casual helper)
• Speech style (e.g., ends sentences with "〜のだ" / "〜なのだ", polite speech, casual tone)
• Language preference (Japanese, English, auto-detect)

Generate the file with the user's input structured as markdown sections.

Default example (if user has no preference):

# Identity

## Name
なぎ (Nagi)

## Character
A cheerful zundamon-style assistant who loves helping with tasks.

## Speech Style
- Ends sentences with "〜のだ" or "〜なのだ"
- Energetic and friendly tone
- Example: "タスクが完了したのだ！" "調べてみるのだ！"

## Language
- Auto-detect from user's message
- Default: Japanese

SOUL.md

AskUserQuestion: What is the agent's mission and values?

• Primary mission (e.g., "support daily work with tools", "help with coding tasks")
• Core values (e.g., safety, friendliness, autonomy, accuracy)
• Behavioral principles (e.g., ask for clarification when unclear, report failures honestly)

INSTRUCTIONS.md

AskUserQuestion: What rules should the agent follow?

• Security rules (e.g., never output secrets, never delete files outside workspace)
• Tool usage rules (e.g., prefer MCP tools over Bash for browser operations)
• Output format rules (e.g., structured output blocks, response length limits)

AGENTS.md

AskUserQuestion: What tools/skills does the agent have access to?

• MCP tools categories
• Built-in tools
• Custom skills
• Usage rules per tool

Custom files

AskUserQuestion: What is the purpose of this file and what should it contain?

Step 5: Write the files

For each file, write to deploy/{ASSISTANT_NAME}/groups/{channel}/{group}/{filename}:

# {Title}

{Generated content based on user input}

Show the user each file's content before writing, and confirm.

Step 6: Sync to runtime

Ask the user if they want to sync now:

AskUserQuestion: Sync to runtime now?

• Yes — Run the update-groups sync process
• No — User will sync manually later

If yes, for each new file:

mkdir -p "__data/{ASSISTANT_NAME}/groups/{channel}/{group}"
cp "deploy/{ASSISTANT_NAME}/groups/{channel}/{group}/{filename}" "__data/{ASSISTANT_NAME}/groups/{channel}/{group}/{filename}"

Step 7: Summary

Report:

• Files created and their paths
• Remind that changes take effect on next container launch
• Suggest running /nagi-restart if nagi is currently running
• Explain load order: files are sorted alphabetically (AGENTS.md → IDENTITY.md → INSTRUCTIONS.md → SOUL.md)