dokugent agent¶
Scaffold a new agent identity and its associated metadata.
The agent command initializes an agent using an interactive CLI wizard. It captures key identity fields and produces metadata artifacts used across Dokugent workflows.
How to Use¶
Interactive Wizard¶
Launches the full guided agent setup wizard.
Instant Scaffold ( --t )¶
Skips the wizard and creates a basic identity.json, with optional ecosystem preset.
Agent Identity Fields¶
During interactive mode, you’ll be asked to provide:
| Field | Description |
|---|---|
agentName | Name of the agent |
description | What the agent is expected to do |
roles | Functional roles (e.g., summarizer, formatter) |
customRole | Optional custom role, if "other" is selected |
processableTypes | Accepted input types (e.g., "english, markdown structure") |
owner | Name of the person or org who owns the agent |
ownerId | ID or domain of the owner |
mainTask | The agent's core instruction or behavior summary |
Flags and Parameters¶
| Flag | Description |
|---|---|
--agent | Specify an agent slug to show, edit, or check. |
--edit | Open the agent’s identity.json in your default editor. |
--show | Print current identity metadata to console. |
--check | Validate the identity.json (step 1 of 6). |
--t | Generate a scaffolded agent identity (skips wizard). Use with --e to include ecosystem presets. |
--e <name> | Specify ecosystem to pull default planning files. |
--ls | List all agents, showing which is current and latest. |
--use <slug> | Set an agent as the current one (via symlink). |
Agent Commands¶
Create a new agent (wizard)¶
Create agent using ecosystem preset¶
List agents¶
Shows agent folders and highlights 🟢 current and 🕓 latest.
Set active agent¶
Updates the current symlink used by all other commands.
View agent metadata¶
Edit agent metadata¶
Validate identity.json¶
Ecosystem Templates¶
Presets scaffold agents based on AI company planning styles. Used with --e.
-
nvidia— Optimized for multimodal video/audio agents using RAG + vision pipelines. ↳ NVIDIA AI Blueprint for Video Analytics -
openai— For agents that use GPT-style function calling or plugin architectures. ↳ OpenAI Function Calling Docs -
anthropic— Constitutional AI-style scaffolds with dialog-style instructions & critique roles. ↳ Anthropic Constitutional AI Paper -
google— Gemini-style multimodal analysis with search-centric planning. ↳ Google DeepMind Gemini Overview -
microsoft— Microsoft 365/Autogen-style agents for productivity environments. ↳ Microsoft Autogen Framework
Each template sets up default plan.md, tools.md, criteria.md, and mcp-schema.md to match the selected ecosystem.
Agent Folder Structure¶
A new agent is created at:
Contents:
identity.json— agent profile- additional template files from ecosystem (if any)
⚠️ If you rename
agentName, rename the folder too.
Workflow Guide¶
- Create an agent:
dokugent agentor--t - Validate it:
--check - Begin planning:
dokugent plan - Link tools and criteria
- Certify the agent
Best Practices¶
- Use descriptive, lowercase
agentNamevalues - Align each agent with one main task
- Use
--eonly when needed — don’t over-scaffold - Validate (
--check) before publishing or certifying