# Hugin Scaffold > Generate starter files for a new Hugin agent Use when this capability is needed. - Skill: `tomevault-io/hugin-scaffold` (Agent Skill, multi-file: 2 files) - Install (CLI): `npx skillmds@latest add tomevault-io/hugin-scaffold` - Raw SKILL.md: https://api.skillmd.com/api/skills/tomevault-io/hugin-scaffold/raw - Safety review: pending (external: skill-scanner PASS, skillspector PASS) - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: AI & ML - Author: tomevault-io (https://skillmd.com/u/tomevault-io) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/tomevault-io/hugin-scaffold --- # Hugin Agent Scaffold Generator Generate starter files for a new Hugin agent by reading templates from this plugin and customizing them. ## Arguments Parse the user's arguments to extract: 1. **agent_name**: Name for the agent (required, snake_case) 2. **type**: One of `minimal`, `tool`, or `pipeline` (default: `minimal`) Example invocations: - `/hugin-agent-creator:hugin-scaffold my_agent` - Creates minimal agent - `/hugin-agent-creator:hugin-scaffold data_processor tool` - Creates agent with custom tool - `/hugin-agent-creator:hugin-scaffold report_generator pipeline` - Creates multi-stage pipeline ## Implementation Steps ### Step 1: Parse Arguments Extract `agent_name` and `type` from user input. Validate `agent_name` is snake_case. ### Step 2: Create Directory Structure Use Bash to create the directories: ```bash mkdir -p [agent_name]/configs [agent_name]/tasks [agent_name]/templates # For 'tool' type also: mkdir -p [agent_name]/tools ``` ### Step 3: Read and Customize Templates **IMPORTANT**: Read the actual template files from this plugin, then customize them. #### For ALL types, read these templates: 1. **Read** `templates/minimal-config.yaml` from this plugin - Replace `my_agent` with `[agent_name]` - Replace `my_system` with `[agent_name]_system` - For 'tool' type: add the custom tool to the tools list - Write to `[agent_name]/configs/[agent_name].yaml` 2. **Read** `templates/minimal-task.yaml` from this plugin - Replace `my_task` with `[agent_name]_task` - Write to `[agent_name]/tasks/[agent_name]_task.yaml` 3. **Read** `templates/minimal-template.yaml` from this plugin - Replace `my_system` with `[agent_name]_system` - Write to `[agent_name]/templates/[agent_name]_system.yaml` #### For 'tool' type, also read: 4. **Read** `templates/tool-definition.yaml` from this plugin - Replace `my_tool` with `[agent_name]_tool` - Write to `[agent_name]/tools/[agent_name]_tool.yaml` 5. **Read** `templates/tool-implementation.py` from this plugin - Replace `my_tool` with `[agent_name]_tool` - Write to `[agent_name]/tools/[agent_name]_tool.py` #### For 'pipeline' type: Instead of the single task, create 3 stage tasks: 1. **Read** `templates/minimal-task.yaml` as base, then create: **tasks/stage_1_extract.yaml:** ```yaml name: stage_1_extract description: Extract data from input (Stage 1) parameters: raw_input: type: string description: Raw input data required: false default: "Sample input data" task_sequence: - stage_2_transform - stage_3_output pass_result_as: extracted_data prompt: | STAGE 1: EXTRACT Raw Input: {{ raw_input.value }} Your task: 1. Parse the raw input 2. Extract key information 3. Use finish with your structured extraction as the result The result will be passed to Stage 2. ``` **tasks/stage_2_transform.yaml:** ```yaml name: stage_2_transform description: Transform extracted data (Stage 2) parameters: extracted_data: type: string description: Data from Stage 1 required: false default: "" pass_result_as: transformed_data prompt: | STAGE 2: TRANSFORM Extracted Data: {{ extracted_data.value }} Your task: 1. Transform the extracted data 2. Apply any necessary processing 3. Use finish with your transformed result The result will be passed to Stage 3. ``` **tasks/stage_3_output.yaml:** ```yaml name: stage_3_output description: Generate final output (Stage 3) parameters: transformed_data: type: string description: Data from Stage 2 required: false default: "" prompt: | STAGE 3: OUTPUT Transformed Data: {{ transformed_data.value }} Your task: 1. Generate the final output 2. Format it appropriately 3. Use finish with your final result ``` 2. Update the system template for pipeline context: ```yaml name: [agent_name]_system template: | You are a data processing assistant working in a multi-stage pipeline. Follow your stage instructions carefully. Your output will be passed to the next stage. Always use the finish tool when your stage is complete, including your result. ``` ### Step 4: Print Success Message After creating all files, print: ``` Created [agent_name] agent ([type] type) Directory: ./[agent_name]/ Files created: - configs/[agent_name].yaml - tasks/[task_name].yaml - templates/[agent_name]_system.yaml [If tool type:] - tools/[agent_name]_tool.yaml - tools/[agent_name]_tool.py Run your agent: uv run hugin run --task [task_name] --task-path ./[agent_name] Next steps: 1. Edit configs/[agent_name].yaml to customize the agent 2. Edit tasks/[task_name].yaml to define your task logic 3. Edit templates/[agent_name]_system.yaml for the system prompt [If tool type:] 4. Implement your tool in tools/[agent_name]_tool.py [If pipeline type:] 4. Customize each stage task in tasks/ For more guidance: /hugin-agent-creator:hugin-guide ``` ## Template File Locations All templates are in this plugin's `templates/` directory: - `templates/minimal-config.yaml` - `templates/minimal-task.yaml` - `templates/minimal-template.yaml` - `templates/tool-definition.yaml` - `templates/tool-implementation.py` **Always read these files** rather than hardcoding their content. This ensures scaffolded agents use the latest template versions. --- > Converted and distributed by [TomeVault](https://tomevault.io/claim/gimlelabs) — claim your Tome and manage your conversions.