Newstaff

---
name: newstaff
description: >-
  A skill for hiring and creating a new Digital Anima in the AnimaWorks organization.
  Based on an interview, create a character sheet (Markdown) and use the CLI command (animaworks anima create) to
  identity/injection/permissionsgenerate everything at once. After creation, self-maintain via bootstrap.
  "Create a new employee" "Hire someone" "New employee" "Hiring" "Create Anima" "Recruit" "Add a team member" "Add personnel" "Create a subordinate" "Add staff" "hire" "recruit" "team member"
---


# Skill: Hiring a New Employee

## Prerequisites

- The direction of the new employee's role must be decided (if unclear, conduct an interview)

## Procedure

### 1. Interview (minimal is fine)

Interview the requester for the following information. **Only the bold items are required**; if others are unspecified, generate them automatically:

**Required:**
- **English name** (lowercase alphanumeric only. This becomes the directory name)
- **Role/specialty**: What they will handle (e.g., research, development, communication, infrastructure monitoring)
- **Personality direction**: Temperature rather than function (e.g., energetic, gal, airheaded, passionate, ojou, older sister). If the organization already has two cool types, don't choose that
- **Face type**: Choose one from `{data_dir}/prompts/face_types.md`. Avoid excessive overlap with existing members

**Optional (reflect if specified, auto-generate if not):**
- Japanese name
- Age
- Any other preferences

**Technical configuration (use defaults if not specified):**
- Role: `commander` (can delegate to other employees) or `worker` (receives delegation)
- supervisor: The English name of the Anima who will be the supervisor (required for worker; if unspecified, self)

**Brain (LLM model) configuration:**

Present the following table and have them choose:

| Level | Execution mode | Example models | Features | credential |
|--------|-----------|-------------|------|------------|
| S | autonomous | `claude-opus-4-6`, `claude-sonnet-4-6` | Claude Agent SDK. Most capable | anthropic |
| A | autonomous | `openai/gpt-4.1`, `google/gemini-2.5-pro`, `vertex_ai/gemini-2.5-flash` | Via LiteLLM. Tool use available | openai / google / azure / vertex |
| B | assisted | `ollama/gemma3:27b`, `ollama/qwen2.5-coder:32b` | No tools. Local execution, low cost | ollama |

※ If not specified, use the default (claude-sonnet-4 / autonomous / anthropic).

### 2. Character design (auto-generated)

Flesh out the character from the information gathered in the interview. **Personality comes first, function comes second.**

Always Read the following in this order:
1. `{data_dir}/prompts/face_types.md`
2. `{data_dir}/prompts/character_design_guide.md`

Don't associate a cool beauty with the role. Don't make hobbies an extension of work. Check that hair color, face type, and speech style don't overlap with existing members.

### 3. Create a character sheet and generate everything at once via CLI

Following the **character sheet specification**, write the results of the interview and design to a character sheet file, then create via CLI command:

1. Write the character sheet to a file (e.g., `/tmp/{english_name}.md`)
2. Execute the following command:

```bash
animaworks anima create --from-md /tmp/{english_name}.md --name {english_name} --supervisor {supervisor_english_name}
```

**supervisor configuration:**
- Specify explicitly with the `supervisor` parameter (recommended)
- If omitted: retrieved from the `| 上司 |` field of the character sheet
- If neither: self (the calling Anima) becomes the supervisor

**Character sheet specification:**

```markdown
# キャラクターシート: {Japanese name}

## 基本情報

| 項目 | 設定 |
|------|------|
| 英名 | {lowercase alphanumeric} |
| 日本語名 | {Japanese full name} |
| 役職/専門 | {Role description} |
| 上司 | {supervisor English name} |
| 役割 | {commander / worker} |
| 実行モード | {autonomous / assisted} |
| モデル | {model name} |
| credential | {anthropic / openai / google / ollama} |

## 人格 (→ identity.md)

{Personality, speaking style, values, backstory, appearance, etc.}

## 役割・行動方針 (→ injection.md)

{Responsible areas, decision criteria, reporting rules, conduct standards, etc.}

## 権限 (→ permissions.json) [省略可]

{If omitted: default template applied}

## 定期業務 (→ heartbeat.md, cron.md) [省略可]

{If omitted: generic template applied. New Anima self-adjusts in bootstrap}

## 初回起動指示 (→ bootstrap.md 追加指示) [省略可]

{If omitted: standard bootstrap only}
```

**Required sections**: Basic information, personality, role and action policy
**Optional sections**: Permission, periodic tasks, initial startup instruction

This automatically executes the following:
- Batch creation of the directory structure
- Placement of skeleton files
- Placement of bootstrap.md
- Creation of status.json (including supervisor)
- Registration in config.json (model, supervisor, etc.)
- Applying defaults to omitted sections

### 4. Confirm the model configuration in config.json

`animaworks anima create` automatically registers in config.json, but confirm and supplement the following:

- `model`: The model name decided during the interview
- `credential`: The credential name to use
- `execution_mode`: autonomous or assisted
- `speciality`: Job title / specialty

### 4.5. Propose and configure heartbeat frequency

Based on the new employee's work drive type, propose a periodic heartbeat interval to the requester and configure the agreed value:

| Drive type | Recommended interval | Example |
|--------|---------|-----|
| Spontaneous situational judgment and coordination as main work | 30 minutes (default) | Team coordinator, progress supervisor |
| Message/event-driven as the main driver | 60–120 minutes | Review lead, development lead |
| Cron periodic tasks as the main driver | 120–240 minutes | Monitoring, accounting, legal |

Add `"heartbeat_interval_minutes": <分>` (1–1440) to `{data_dir}/animas/{英名}/status.json` for the configuration.
Message-triggered heartbeats and cron run independently of this configuration, so extending the interval does not reduce responsiveness. When in doubt, choose a longer interval (empty heartbeats waste tokens).

### 5. Reflect on the server

```
Bash: curl -s -X POST http://localhost:18500/api/system/reload
```

### 6. Report to the requester

Report the completion of the hire:
- The new employee's name and role
- The configured technical stack (model, execution mode)

⚠️ Do not report avatar image generation (the new Anima itself generates it via bootstrap)

### From here on, the new Anima executes autonomously:
- Enriching identity.md / injection.md
- Self-designing heartbeat.md / cron.md
- Generating the avatar image (with supervisor reference)
- Reporting arrival to the supervisor